Image acquisition model training method, image detection method, device and equipment

Abstract

The application discloses a kind of training method of image acquisition model, image detection method, device and equipment, belong to image processing technical field.Method includes: obtaining label image pair of sample image;According to first network model, the predicted image pair of sample image is obtained, and the second predicted image is the heat map of sample image in Hough space obtained by prediction;Second network model is obtained based on label image pair and predicted image pair adjustment first network model;In response to satisfying training termination condition, second network model is used as image acquisition model.The application can quickly and accurately determine the heat map of image in Hough space, realize accurate image detection result based on the heat map of image in Hough space to determine, to improve the accuracy of subsequent analysis processing.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of image processing, in particular to a training method of an image acquisition model, an image detection method, device and equipment. BACKGROUND

[0002] With the continuous development of computers, the application range of image detection technology is wider and wider, and the types of image detection technology methods are more and more. Hough transformation is one of the methods of image detection technology. Based on the principle of Hough transformation, the planes, straight lines, ellipses and the like of objects in an image can be detected, so that subsequent analysis and processing can be performed.

[0003] Taking the detection of the planes of objects in an image based on the principle of Hough transformation as an example, the principle is to utilize the duality of points and planes, to convert the points in an image space into a plane in a Hough space through a plane expression form, so as to obtain a heat map of the Hough space. By finding the peak points in the Hough space, the target plane in the image space is obtained, i.e. the plane of the object in the image is obtained.

[0004] In the related art, an accumulator is often used to convert each point in an image space into a heat map of a Hough space based on the principle of Hough transformation, so as to determine an image detection result based on the heat map of the Hough space, i.e. to determine the planes, straight lines, ellipses and the like of objects in an image. However, the accumulator is implemented based on a traditional mathematical transformation mode, and needs to constantly iterate parameters, which is very slow and has low accuracy, thereby affecting the accuracy of subsequent analysis and processing. SUMMARY

[0005] Embodiments of the present application provide a training method of an image acquisition model, an image detection method, device and equipment, which can be used to solve the problem of low accuracy of a heat map of an image in a Hough space in the related art. The technical solution comprises the following content based on the present application.

[0006] In one aspect, the present application provides a training method of an image acquisition model, comprising:

[0007] obtaining a label image pair of a sample image, the label image pair comprising a first label image and a second label image, the first label image being a heat map of the sample image in an image space obtained by labeling, and the second label image being a heat map of the sample image in a Hough space obtained by labeling;

[0008] obtaining a predicted image pair of the sample image according to a first network model, the predicted image pair comprising a first predicted image and a second predicted image, the first predicted image being a heat map of the sample image in the image space obtained by prediction, and the second predicted image being a heat map of the sample image in the Hough space obtained by prediction;

[0009] adjust the first network model based on the label image pair and the predicted image pair to obtain a second network model;

[0010] In response to meeting a training termination condition, the second network model is taken as an image acquisition model.

[0011] In another aspect, an embodiment of the present application provides an image detection method, which comprises:

[0012] obtaining a to-be-detected image;

[0013] obtaining a target image of the to-be-detected image according to an image acquisition model, the target image being a heat map of the to-be-detected image in a Hough space, and the image acquisition model being obtained according to the training method of any of the above image acquisition models;

[0014] determining an image detection result of the to-be-detected image based on the target image.

[0015] In another aspect, an embodiment of the present application provides a training device of an image acquisition model, which comprises:

[0016] a first obtaining module configured to obtain a label image pair of a sample image, the label image pair comprising a first label image and a second label image, the first label image being a heat map of the sample image in an image space obtained through labeling, and the second label image being a heat map of the sample image in a Hough space obtained through labeling;

[0017] a second obtaining module configured to obtain a predicted image pair of the sample image according to a first network model, the predicted image pair comprising a first predicted image and a second predicted image, the first predicted image being a heat map of the sample image in the image space obtained through prediction, and the second predicted image being a heat map of the sample image in the Hough space obtained through prediction;

[0018] an adjusting module configured to adjust the first network model based on the label image pair and the predicted image pair to obtain a second network model;

[0019] a determining module configured to, in response to meeting a training termination condition, take the second network model as an image acquisition model.

[0020] In a possible implementation, the second obtaining module is configured to: obtain a first image space feature of the sample image, the first image space feature being used to represent a feature of the sample image in the image space; determine the first predicted image based on the first image space feature; determine a first Hough space feature based on the first image space feature, the first Hough space feature being used to represent a feature of the sample image in the Hough space; and determine the second predicted image based on the first Hough space feature.

[0021] In a possible implementation, the second obtaining module is configured to: rectify the first image space feature to obtain a rectified first image space feature; and determine the first Hough space feature based on the rectified first image space feature.

[0022] In a possible implementation, the adjusting module is configured to: determine a loss value of the first network model based on the label image pair and the predicted image pair; and adjust the first network model based on the loss value of the first network model to obtain a second network model.

[0023] In a possible implementation, the second obtaining module is further configured to: determine a second image space feature based on the first Hough space feature and the first image space feature; determine a third predicted image based on the second image space feature, the third predicted image being a heat map of the sample image in the image space obtained by prediction; determine a second Hough space feature based on the second image space feature; and determine a fourth predicted image based on the second Hough space feature, the fourth predicted image being a heat map of the sample image in the Hough space obtained by prediction.

[0024] The adjusting module is configured to adjust the first network model based on the label image pair, the predicted image pair, the third predicted image, and the fourth predicted image to obtain a second network model.

[0025] In a possible implementation, the second obtaining module is configured to: determine a third image space feature based on the first Hough space feature; fuse the first image space feature and the third image space feature to obtain a first fused feature; and determine the second image space feature based on the first fused feature.

[0026] In a possible implementation, the adjusting module is configured to: obtain a first loss value according to the label image pair and the predicted image pair; obtain a second loss value according to the label image pair, a third predicted image, and a fourth predicted image; obtain a loss value of the first network model based on the first loss value and the second loss value; and adjust the first network model based on the loss value of the first network model to obtain a second network model.

[0027] In a possible implementation, the first image space feature includes at least two sub-image space features.

[0028] The second obtaining module is configured to determine the first predicted image based on the at least two sub-image space features.

[0029] In a possible implementation, the second obtaining module is configured to, for any one of the at least two sub-image space features, determine a sub-Hough space feature corresponding to the any one of the sub-image space features, and the first Hough space feature includes the sub-Hough space feature corresponding to each of the sub-image space features; and determine the second predicted image based on the sub-Hough space feature corresponding to each of the sub-image space features.

[0030] In a possible implementation, the second obtaining module is configured to, in response to the any one of the sub-image space features being any one of the at least two sub-image space features except for a first sub-image space feature, determine a second fusion feature corresponding to the any one of the sub-image space features based on the any one of the sub-image space features and a previous sub-image space feature of the any one of the sub-image space features; and determine the sub-Hough space feature corresponding to the any one of the sub-image space features based on the second fusion feature corresponding to the any one of the sub-image space features and the previous sub-image space feature of the any one of the sub-image space features.

[0031] In a possible implementation, the second obtaining module is configured to determine a second fusion feature corresponding to a previous sub-image space feature of the any one of the sub-image space features based on the previous sub-image space feature; determine a fourth image space feature based on the second fusion feature corresponding to the previous sub-image space feature; determine a third Hough space feature based on the fourth image space feature; and determine the second fusion feature corresponding to the any one of the sub-image space features based on the any one of the sub-image space features, the third Hough space feature, and the second fusion feature corresponding to the previous sub-image space feature.

[0032] In a possible implementation, the second obtaining module is configured to determine a sub-Hough space feature corresponding to a previous sub-image space feature of the any one sub-image space feature based on the previous sub-image space feature; determine a sub-Hough space feature corresponding to the second fusion feature based on the second fusion feature corresponding to the any one sub-image space feature; and determine a sub-Hough space feature corresponding to the any one sub-image space feature based on the sub-Hough space feature corresponding to the second fusion feature and the sub-Hough space feature corresponding to the previous sub-image space feature.

[0033] In another aspect, an image detection device is provided, and the device comprises:

[0034] A first obtaining module is configured to obtain a to-be-detected image.

[0035] A second obtaining module is configured to obtain a target image of the to-be-detected image according to an image obtaining model, the target image being a heat map of the to-be-detected image in a Hough space, and the image obtaining model being obtained according to the training method of any one of the above image obtaining models.

[0036] A determining module is configured to determine an image detection result of the to-be-detected image based on the target image.

[0037] In a possible implementation, the to-be-detected image is a brain image, and the second obtaining module is configured to obtain a first image space feature of the brain image according to an image obtaining model, the first image space feature of the brain image being used to represent a feature of the brain image in an image space; determine a first Hough space feature of the brain image based on the first image space feature of the brain image, the first Hough space feature of the brain image being used to represent a feature of the brain image in the Hough space; and determine a target image of the brain image based on the first Hough space feature of the brain image.

[0038] In a possible implementation, the second obtaining module is configured to determine a second image space feature of the brain image based on the first Hough space feature of the brain image and the first image space feature of the brain image; determine a second Hough space feature of the brain image based on the second image space feature of the brain image; and determine a target image of the brain image based on the second Hough space feature of the brain image.

[0039] In a possible implementation, the determining module is configured to perform a non-maximum suppression process on the target image to obtain at least one peak point in the target image; and determine an image detection result of the to-be-detected image based on the at least one peak point in the target image.

[0040] In another aspect, an electronic device is provided. The electronic device includes a processor and a memory. The memory stores at least one program code. The at least one program code is loaded and executed by the processor to cause the electronic device to implement any of the above-described image acquisition model training methods or any of the above-described image detection methods.

[0041] In another aspect, a computer-readable storage medium is provided. The computer-readable storage medium stores at least one program code. The at least one program code is loaded and executed by a processor to cause a computer to implement any of the above-described image acquisition model training methods or any of the above-described image detection methods.

[0042] In another aspect, a computer program or computer program product is provided. The computer program or computer program product stores at least one computer instruction. The at least one computer instruction is loaded and executed by a processor to cause a computer to implement any of the above-described image acquisition model training methods or any of the above-described image detection methods.

[0043] The technical solutions provided by the embodiments of the present application at least bring the following beneficial effects:

[0044] The technical solutions provided by the embodiments of the present application are based on the heat map of the sample image in the Hough space and the heat map of the sample image in the image space to train the image acquisition model. The image acquisition model learns the key point features in the Hough space and the semantic features in the image space, can quickly and accurately determine the heat map of the image in the Hough space, accurately determine the image detection result based on the heat map of the image in the Hough space, and improve the accuracy of subsequent analysis and processing. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0046] Figure 1 is an implementation environment schematic diagram of an image acquisition model training method or an image detection method provided by the embodiments of the present application;

[0047] Figure 2 is a flowchart of an image acquisition model training method provided by the embodiments of the present application;

[0048] Figure 3 is an image schematic diagram provided by the embodiments of the present application;

[0049] Figure 4 is another image schematic diagram provided by an embodiment of the present application;

[0050] Figure 5 is still another image schematic diagram provided by an embodiment of the present application;

[0051] Figure 6 is a schematic diagram of a fast deep Hough transform network provided by an embodiment of the present application;

[0052] Figure 7 is a schematic diagram of a reverse deep Hough transform provided by an embodiment of the present application;

[0053] Figure 8 is a schematic diagram of a first network model provided by an embodiment of the present application;

[0054] Figure 9 is a determination schematic diagram of a second fusion feature corresponding to a feature of any sub-image space provided by an embodiment of the present application;

[0055] Figure 10 is another schematic diagram of a first network model provided by an embodiment of the present application;

[0056] Figure 11 is a flowchart of an image detection method provided by an embodiment of the present application;

[0057] Figure 12 is a processing schematic diagram of a brain scan image provided by an embodiment of the present application;

[0058] Figure 13 is a processing schematic diagram of a fetal scan image provided by an embodiment of the present application;

[0059] Figure 14 is a processing schematic diagram of a cell image provided by an embodiment of the present application;

[0060] Figure 15 is a processing schematic diagram of a photographic image provided by an embodiment of the present application;

[0061] Figure 16 is a structural schematic diagram of a training device of an image acquisition model provided by an embodiment of the present application;

[0062] Figure 17 is a structural schematic diagram of an image detection device provided by an embodiment of the present application;

[0063] Figure 18 is a structural schematic diagram of a terminal device provided by an embodiment of the present application;

[0064] Figure 19is a structural schematic diagram of a server provided by an embodiment of the present application;

[0065] Figure 20 is a schematic diagram of an image provided by an embodiment of the present application;

[0066] Figure 21 is another schematic diagram of an image provided by an embodiment of the present application;

[0067] Figure 22 is still another schematic diagram of an image provided by an embodiment of the present application;

[0068] Figure 23 is a schematic diagram of a reverse depth Hough transform provided by an embodiment of the present application. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solutions and advantages of the present application more clear, the embodiments of the present application will be further described in detail below with reference to the drawings.

[0070] First, the terms involved in the embodiments of the present application are explained.

[0071] Image Space is a three-dimensional geometric space of an image.

[0072] Hough Space is a polar coordinate space (also called a parameter space) of an image.

[0073] Hough Transform is a feature extraction method, which, given a straight line, an ellipse, a plane or the like equation, performs voting in a polar coordinate space, and through detecting local peak points of the accumulated polar coordinate space, locates and detects geometric patterns. According to different equations, Hough Transform can be divided into Line Hough Transform (LHT), Plane Hough Transform (PHT) and Elliptic Hough Transform (EHT).

[0074] Deep Hough Transform (DHT) networks are convolutional neural networks (CNNs) that integrate the Hough Transform method. Using DHT networks, the features of a sample image in image space are determined based on its features in image space, resulting in a heatmap of the sample image in Hough space. DHT networks are divided into 3D Deep Hough Transform (3D DHT) networks and 2D Deep Hough Transform (2D DHT) networks. 3D DHT networks are used to process 3D sample images, while 2D DHT networks are used to process 2D sample images.

[0075] The Inverse Deep Hough Transform (IDHT) network is a CNN that integrates the Inverse Hough Transform method. Using the IDHT network, the features of a sample image in the image space are determined based on its features in the Hough space, thus obtaining a heatmap of the sample image in the image space. IDHT networks are divided into 3D Inverse Deep Hough Transform (3D IDHT) networks and 2D Inverse Deep Hough Transform (2D IDHT) networks. The 3D IDHT network is used to process 3D sample images, and the 2D IDHT network is used to process 2D sample images.

[0076] Figure 1 This is a schematic diagram illustrating the implementation environment of a training method for an image acquisition model or an image detection method provided in an embodiment of this application, such as... Figure 1 The implementation environment shown includes an electronic device 11. The training method or image detection method of the image acquisition model in this embodiment can be executed by the electronic device 11. Exemplarily, the electronic device 11 may include at least one of a terminal device or a server.

[0077] The terminal device can be at least one of a smart phone, a game console, a desktop computer, a tablet computer, an e-book reader, an MP3 (Moving Picture Experts Group Audio Layer III) player, an MP4 (Moving Picture Experts Group Audio Layer IV) player, and a laptop computer.

[0078] The server can be a single server, or a server cluster composed of multiple servers, or any one of a cloud computing platform and a virtualization center, and the embodiments of the present application do not limit this. The server can be in communication connection with the terminal device through a wired network or a wireless network. The server can have functions of data processing, data storage, and data transceiving, and the embodiments of the present application do not limit this.

[0079] The technical solutions of the optional embodiments of the present application are implemented based on artificial intelligence technology. Artificial intelligence (AI) is the theory, method, technology and application system for using a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which aims to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0080] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, automatic driving, intelligent transportation, etc.

[0081] Computer Vision (CV) is a science that studies how to make machines "see". More specifically, it refers to using cameras and computers to replace human eyes to identify, detect and measure targets, and further perform image processing to make computer processing more suitable for human observation or transmission to instrument detection. As a scientific discipline, computer vision researches related theories and technologies, and attempts to establish artificial intelligence systems that can obtain information from images or multi-dimensional data. Computer vision technology generally includes image processing, image recognition, image semantic understanding, image retrieval, Optical Character Recognition (OCR), video processing, video semantic understanding, video content / behavior recognition, three-dimensional object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, autonomous driving, intelligent transportation, and other technologies. It also includes common face recognition, fingerprint recognition and other biometric identification technologies.

[0082] With the research and progress of artificial intelligence technology, artificial intelligence technology is being researched and applied in many fields, such as common smart home, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned vehicles, autonomous vehicles, drones, robots, smart medical care, smart customer service, Internet of Vehicles, autonomous driving, intelligent transportation, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0083] The scheme provided by the embodiments of the present application relates to computer vision and other technologies of artificial intelligence, which is specifically explained by the following embodiments.

[0084] Based on the above implementation environment, the embodiments of the present application provide a training method of an image acquisition model, which is used to Figure 2 The flowchart of the training method of the image acquisition model provided by the embodiments of the present application is taken as an example, which can be executed by the electronic device 11 in Figure 1 As shown in Figure 2 The method includes steps 201 to 204.

[0085] Step 201, obtaining a label image pair of a sample image.

[0086] The label image pair includes a first label image and a second label image. The first label image is a heat map of the sample image in image space obtained by labeling, and the second label image is a heat map of the sample image in Hough space obtained by labeling.

[0087] The embodiments of the present application do not limit the sample image. For example, the sample image includes but is not limited to brain image, landscape image, road image, fetal image, cell image, etc. The number of sample images is multiple.

[0088] The label image pair of the sample image comprises a first label image and a second label image. The first label image is a heat map of the sample image in the image space, and a value of any pixel point in the first label image represents a probability of the pixel point being on the target straight line or on the target ellipse or on the target plane. The second label image is a heat map of the sample image in the Hough space, and a value of any pixel point in the second label image represents a probability of the pixel point being the target straight line or the target ellipse or the target plane.

[0089] Please refer to Figure 3 or Figure 20 , Figure 3 or Figure 20 is an image schematic diagram provided by an embodiment of the present application. The image shown by reference numeral 301 is a sample image. The heat map of the sample image 301 in the image space is shown by reference numeral 302, and the heat map of the sample image 301 in the Hough space is shown by reference numeral 303. The effect diagram corresponding to reference numeral 303 is shown by reference numeral 2002. The image shown by reference numeral 304 is an image obtained by superimposing the sample image 301 and the heat map 302, and the effect diagram corresponding to reference numeral 304 is shown by reference numeral 2001.

[0090] Please refer to Figure 4 or Figure 21 , Figure 4 or Figure 21 is another image schematic diagram provided by an embodiment of the present application. In actual situations, the heat map of the sample image 301 in the image space is a strip. The image shown by reference numeral 401 is an image obtained by superimposing the sample image and the heat map of the sample image in the image space, and the effect diagram corresponding to reference numeral 401 is shown by reference numeral 2101. By comparing the image shown by reference numeral 304 and the image shown by reference numeral 401, it can be obviously seen that, in the image shown by reference numeral 304, the heat map of the sample image 301 in the image space is a straight line, and in the image shown by reference numeral 401, the heat map of the sample image 301 in the image space is a strip. The images shown by reference numerals 402 and 403 are both images obtained by superimposing the sample image and the heat map of the sample image in the image space, the effect diagram corresponding to reference numeral 402 is shown by reference numeral 2102, and the effect diagram corresponding to reference numeral 403 is shown by reference numeral 2103. The three images shown by reference numerals 401, 402 and 403 are images of three different perspectives.

[0091] Please refer to Figure 5 or Figure 22 , Figure 5 or Figure 22The three images shown by reference sign 501 are images obtained by superimposing the heat map of the sample image in the image space and the sample image under different viewing angles, and the effect diagram corresponding to reference sign 501 is shown in reference sign 2201. The three images shown by reference sign 502 are images obtained by superimposing the heat map of the sample image in the image space and the sample image under different viewing angles, and the effect diagram corresponding to reference sign 502 is shown in reference sign 2202.

[0092] In step 202, a predicted image pair of the sample image is obtained according to the first network model.

[0093] The predicted image pair includes a first predicted image and a second predicted image. The first predicted image is a heat map of the sample image in the image space obtained by prediction, and the second predicted image is a heat map of the sample image in the Hough space obtained by prediction.

[0094] In the embodiments of the present application, the sample image is input into the first network model, and the first network model outputs the first predicted image and the second predicted image. The first predicted image is a heat map of the sample image in the image space, and the value of any pixel point in the first predicted image represents the probability of the pixel point on the target straight line or on the target ellipse or on the target plane. The second predicted image is a heat map of the sample image in the Hough space, and the value of any pixel point in the second predicted image represents the probability of the pixel point being the target straight line or the target ellipse or the target plane.

[0095] In step 203, the first network model is adjusted based on the label image pair and the predicted image pair to obtain a second network model.

[0096] In the embodiments of the present application, the loss value of the first network model is calculated based on the label image pair and the predicted image pair, the model parameters of the first network model are adjusted based on the loss value of the first network model, and the second network model is obtained.

[0097] In step 204, the second network model is taken as the image acquisition model in response to satisfying a training termination condition.

[0098] The embodiments of the present application do not limit the training termination condition, and exemplarily, the training termination condition is that the target training number is reached. The value of the target training number can be flexibly set according to artificial experience or actual scene, for example, the target training number is 200.

[0099] In response to satisfying the training termination condition, the second network model is taken as the image acquisition model. In response to not satisfying the training termination condition, the second network model is trained based on the sample image pair in the manner of steps 202 and 203 until the training termination condition is satisfied, and the image acquisition model is obtained.

[0100] The sample image in the above method itself represents semantic information of the image, therefore, the heat map of the sample image in the image space represents semantic features of the image space, and since lines and planes in the image space correspond to points in the Hough space, the heat map of the sample image in the Hough space represents key point features of the Hough space. The image acquisition model is trained based on the heat map of the sample image in the Hough space and the heat map of the sample image in the image space, so that the image acquisition model learns the key point features of the Hough space and the semantic features of the image space, can quickly and accurately determine the heat map of the image in the Hough space, accurately determine the image detection result based on the heat map of the image in the Hough space, and thus improve the accuracy of subsequent analysis and processing.

[0101] In a possible implementation, the method for obtaining the predicted image of the sample image comprises: obtaining a first image space feature of the sample image, the first image space feature being used to represent features of the sample image in an image space; determining a first predicted image based on the first image space feature; determining a first Hough space feature based on the first image space feature, the first Hough space feature being used to represent features of the sample image in a Hough space; and determining a second predicted image based on the first Hough space feature.

[0102] In the embodiments of the present application, the first image space feature of the sample image is extracted according to the first network model. Then, on the one hand, the first predicted image is obtained by upsampling processing based on the first image space feature according to the first network model, and on the other hand, the first Hough space feature is obtained by Hough transform processing based on the first image space feature according to the first network model. Then, the second predicted image is obtained by upsampling processing based on the first Hough space feature according to the first network model.

[0103] The first network model comprises a DHT network, which is used to perform Hough transform processing on the first image space feature to obtain the first Hough space feature. When the sample image is a two-dimensional sample image, the DHT network is a 2D DHT network, and when the sample image is a three-dimensional sample image, the DHT network is a 3D DHT network. The operation principle of the 3D DHT network will be described in detail below.

[0104] In the embodiments of the present application, when the sample image is a three-dimensional sample image, the image acquisition model mainly detects a certain plane (such as a brain midline plane, a standard plane of a fetus, etc.) of the three-dimensional sample image. The expression of the plane in a rectangular coordinate system is Ax+By+Cz+D=0, which can be converted into an expression in a polar coordinate system: The features of the sample image in the image space are R H×W×D , and the features of the sample image in the Hough space are where A, B, C and D are four parameters, x, y and z are three variables, θ and ρ are three parameters, R is a real number, H is height, W is width, and D is depth. The Hough transform is to convert any point in the image space into a plane in the Hough space, and in this way, each point in the image space is converted into a plane in the Hough space, and each plane in the Hough space is accumulated to obtain a Hough key point heat map (i.e. a heat map of the image in the Hough space). The maximum point in the Hough key point heat map corresponds to the key point in the image, and the key point in the image corresponds to the maximum point in the Hough key point heat map. (i.e. phi), θ (i.e. theta) and ρ (i.e. rho). Based on the above information, the 3D DHT network implements the Hough transform processing according to implementation mode A1 and implementation mode A2.

[0105] In the embodiments of the present application, the algorithm of the 3D DHT network includes a forward propagation algorithm of the 3D DHT network (see implementation mode A1) and a backward propagation algorithm of the 3D DHT network (see implementation mode A2).

[0106] Implementation mode A1, forward propagation algorithm of the 3D DHT network.

[0107] Input: volume, / / feature R of the sample image in the image space H×W×D

[0108] phi, / / its value range is [phi_min, phi_max], and the number of samples is n_phi

[0109] theta, / / its value range is [theta_min, theta_max], and the number of samples is n_theta

[0110] rho, / / its value range is [rho_min, rho_max], and the number of samples is n_rho

[0111] f. / / plane equation

[0112] Output: hough. / / feature of the sample image in the Hough space

[0113] The implementation algorithm is as follows:

[0114] function dht3d_forward(volume, hough_range, hough_size) / / Forward propagation algorithm function of 3D DHT network, which includes three inputs: volume, hough_range, and hough_size. hough_range includes the value ranges of phi, theta, and rho, and hough_size includes the sampling numbers of phi, theta, and rho

[0115] / / Create a zero array with size n_phi x n_theta x n_rho as the feature hough of the sample image in Hough space

[0116] for (h, w, d) in mesh_grid(volume_size), / / Create a pixel point for traversing volume, whose coordinates are (h, w, d)

[0117] (x, y, z) <- (h, w, d), / / Transform the coordinates of the pixel point from (h, w, d) to (x, y, z)

[0118] val <- volume[h, w, d], / / Get the pixel value val of the pixel point, which is volume[h, w, d]

[0119] for phi in phi_range, / / Let phi take values in [phi_min, phi_max] in turn

[0120] for theta in theta_range, / / Let theta take values in [theta_min, theta_max] in turn

[0121] rho <- f(x, y, z, theta, phi), / / Substitute x, y, z, theta, phi into the plane equation to calculate rho

[0122] hough[phi, theta, rho] <- val + hough[phi, theta, rho], / / Assign hough[phi, theta, rho] as the sum of val and hough[phi, theta, rho]

[0123] return hough, / / Return the feature hough of the sample image in Hough space

[0124] end function.

[0125] The embodiment of the application can obtain the features of the sample image in the Hough space through the implementation manner A1.

[0126] Implementation manner A2, the back propagation algorithm of the 3D DHT network.

[0127] Input: hough, / / the features of the sample image in the Hough space

[0128] phi, / / the value range of which is [phi_min, phi_max], and the sampling number is n_phi

[0129] theta, / / the value range of which is [theta_min, theta_max], and the sampling number is n_theta

[0130] rho, / / the value range of which is [rho_min, rho_max], and the sampling number is n_rho

[0131] f. / / the plane equation

[0132] Output: volume. / / the features R of the sample image in the image space H×W×D

[0133] The implementation algorithm is as follows:

[0134] function dht3d_backward(hough, hough_range, hough_size) / / the back propagation algorithm function of the 3D DHT network, which includes three inputs of hough, hough_range and hough_size, wherein the hough_range includes the value range of phi, theta and rho, and the hough_size includes the sampling number of phi, theta and rho

[0135] volume <- 0 H×W×D , / / create a zero array with the size of HxWxD as the features volume of the sample image in the image space

[0136] for (h, w, d) in mesh_grid(volume_size), / / create a pixel point for traversing the volume, and the coordinates of the pixel point are (h, w, d)

[0137] accum <- 0, / / obtain the pixel value accum of the pixel point, and the value is 0

[0138] for phi in phi_range, / / let phi take values in [phi_min, phi_max] in turn

[0139] rho <- f(x, y, z, phi, theta), / / substitute x, y, z, theta, phi into the plane equation to calculate rho

[0140] accum <- accum + hough[phi, theta, rho], / / assign accum to the sum of accum and hough[phi, theta, rho]

[0141] volume[h, w, d] <- volume[h, w, d] + accum, / / assign volume[h, w, d] to the sum of volume[h, w, d] and accum

[0142] return volume, / / return the features of the sample image in the image space

[0143] end function.

[0144] The embodiment of the present application can obtain the features of the sample image in the image space through the implementation manner A2.

[0145] In the embodiment of the present application, when the sample image is a two-dimensional sample image, the image acquisition model mainly detects straight lines or ellipses and the like of the two-dimensional sample image. Taking a straight line as an example, the expression of the straight line in the rectangular coordinate system is Ax+By+C=0, which can be converted into the expression in the polar coordinate system: x cos theta+y sin theta=p. The features of the sample image in the image space are R H×W , and the features of the sample image in the Hough space are R θ×ρ . Wherein, A, B and C are three parameters, x and y are two variables, theta and p are two parameters, R is a real number, H is height, and W is width. Based on the above information, the 2D DHT network implements the Hough transform processing.

[0146] The algorithm of the 2D DHT network includes the forward propagation algorithm of the 2D DHT network and the back propagation algorithm of the 2D DHT network. The forward propagation algorithm of the 2D DHT network is similar in principle to the forward propagation algorithm of the 3D DHT network, which is described in detail in the implementation manner A1, and will not be repeated here. The back propagation algorithm of the 2D DHT network is similar in principle to the back propagation algorithm of the 3D DHT network, which is described in detail in the implementation manner A2, and will not be repeated here.

[0147] Optionally, the first Hough space feature is determined based on the first image space feature, including: rectifying the first image space feature to obtain rectified first image space feature; and determining the first Hough space feature based on the rectified first image space feature.

[0148] In the embodiments of the present application, the first network model includes a Fast DHT network, and the Fast DHT network includes a rectification network and a DHT network. The Fast DHT network is used to reduce the amount of calculation and speed up the operation.

[0149] Please refer to Figure 6 , Figure 6 is a schematic diagram of a Fast DHT network provided by the embodiments of the present application. The first image space feature is input into the Fast DHT network, the rectification network rectifies the first image space feature to obtain rectified first image space feature, and the DHT network performs Hough transform on the rectified first image space feature to obtain the first Hough space feature.

[0150] Optionally, the formula of the rectification network is f(x)=max(0,x). Wherein, x is the data of the first image space feature, and f(x) is the data of the rectified first image space feature. Through the rectification network, each data of the first image space feature is detected. When the data is less than 0, the data is reset to 0, and when the data is greater than or equal to 0, the data remains unchanged. In this way, the data 0 in the first image space feature is increased, thereby increasing the sparsity of the first image space feature. The data in the first image space feature corresponds to the pixel points, which is equivalent to increasing the pixel points with the feature of 0. By setting the pixel points with the feature of 0 in the image space not to participate in the voting to the Hough space, the pixel points with the feature of non-0 in the image space are allowed to participate in the voting to the Hough space, the number of pixel points participating in the voting is reduced, and the speed of Hough transform is improved.

[0151] Please refer to Table 1, which is the operation time of the Fast DHT network and the DHT network tested on a display card. In the embodiments of the present application, the operation time of two sample images in the Fast DHT network and the DHT network is tested. The image size corresponding to the heat map of one sample image in the image space is 400x400, and the image size corresponding to the heat map of the sample image in the Hough space is 180x400. The Fast DHT network corresponding to this sample image is the Fast 2D DHT network, and the DHT network corresponding to this sample image is the 2D DHT network. The image size corresponding to the heat map of another sample image in the image space is 192x192x160, and the image size corresponding to the heat map of the sample image in the Hough space is 180x180x160. The Fast DHT network corresponding to this sample image is the Fast 3D DHT network, and the DHT network corresponding to this sample image is the 3D DHT network.

[0152] Table 1

[0153] 2D DHT Fast 2D DHT 3D DHT Fast 3D DHT Time (unit: second) 0.0193 0.0084 26.0 1.3

[0154] As can be seen from Table 1, the operation time of the Fast 2D DHT network is much lower than that of the 2D DHT network, and the operation time of the Fast 3D DHT network is much lower than that of the 3D DHT network. Therefore, the Fast DHT network can reduce the operation amount and speed up the operation.

[0155] In a possible implementation, after determining the first Hough space feature based on the first image space feature, the method further includes: determining a second image space feature based on the first Hough space feature and the first image space feature, the second image space feature being used to represent the feature of the sample image in the image space; determining a third predicted image based on the second image space feature, the third predicted image being a heat map of the sample image in the image space obtained by prediction; determining a second Hough space feature based on the second image space feature, the second Hough space feature being used to represent the feature of the sample image in the Hough space; determining a fourth predicted image based on the second Hough space feature, the fourth predicted image being a heat map of the sample image in the Hough space obtained by prediction; and adjusting the first network model based on the label image pair and the predicted image pair to obtain a second network model, including: adjusting the first network model based on the label image pair, the predicted image pair, the third predicted image and the fourth predicted image to obtain the second network model.

[0156] In the embodiments of the present application, according to the first network model, the second image space feature is determined based on the first image space feature and the first Hough space feature. Then, on the one hand, according to the first network model, the second image space feature is up-sampled to obtain a third predicted image, and on the other hand, according to the first network model, the second image space feature is subjected to Hough transform to obtain a second Hough space feature. Subsequently, according to the first network model, the second Hough space feature is up-sampled to obtain a fourth predicted image.

[0157] Subsequently, a loss value of the first network model is calculated based on the label image pair, the predicted image pair, the third predicted image and the fourth predicted image, and the model parameters of the first network model are adjusted based on the loss value of the first network model to obtain a second network model.

[0158] In the embodiments of the present application, the first predicted image can be obtained based on the first image space feature, and the second predicted image can be obtained based on the first Hough space feature. Subsequently, the second image space feature is determined based on the first image space feature and the first Hough space feature, and the accuracy of the image space feature is improved through further processing of the feature, so that the third predicted image obtained based on the second image space feature is more accurate. Similarly, the accuracy of the second Hough space feature determined based on the second image space feature is improved, so that the fourth predicted image obtained based on the second Hough space feature is more accurate. The first to fourth predicted images are used to adjust the first network model, so that the model learns more information and the convergence speed of the model is accelerated.

[0159] Optionally, the second image space feature is determined based on the first Hough space feature and the first image space feature, including: determining a third image space feature based on the first Hough space feature, the third image space feature being used to represent the feature of the sample image in the image space; fusing the first image space feature and the third image space feature to obtain a first fused feature; and determining the second image space feature based on the first fused feature.

[0160] In the embodiments of the present application, the first network model includes an IDHT network, and the IDHT network performs inverse Hough transform on the first Hough space feature to obtain a third image space feature. The first Hough space feature is obtained based on the first image space feature, that is, the first image space feature is processed to obtain the first Hough space feature, and then the first Hough space feature is processed to obtain the third image space feature, which will cause feature loss. The third image space feature is fused with the first image space feature to reduce the influence caused by the feature loss, to obtain a first fused feature. Subsequently, the second image space feature is determined based on the first fused feature.

[0161] Please refer to Figure 7 orFigure 23 , Figure 7 or Figure 23 is a schematic diagram of a reverse depth Hough transform provided by an embodiment of the present application, the reverse depth Hough transform being used to transform an image from a Hough space to an image space. Among them, the three images shown by the label 701 are heat maps of sample images in the Hough space under different viewing angles, the effect diagram corresponding to the label 701 is shown as the label 2301, the three images shown by the label 702 are heat maps of sample images in the image space under different viewing angles, the effect diagram corresponding to the label 702 is shown as the label 2302, and the Hough space features corresponding to the three images shown by the label 701 are converted into the image space features corresponding to the three images shown by the label 702 through the IDHT network. The three images shown by the label 703 are heat maps of sample images in the Hough space under different viewing angles, the effect diagram corresponding to the label 703 is shown as the label 2303, the three images shown by the label 704 are heat maps of sample images in the image space under different viewing angles, the effect diagram corresponding to the label 704 is shown as the label 2304, and the Hough space features corresponding to the three images shown by the label 703 are converted into the image space features corresponding to the three images shown by the label 704 through the IDHT network.

[0162] It should be noted that the reverse Hough transform processing of the IDHT network is the inverse process of the Hough transform processing of the DHT network. The IDHT network includes a 2D IDHT network and a 3D IDHT network, the 2D IDHT network includes a forward propagation algorithm of the 2D IDHT network and a backward propagation algorithm of the 2D IDHT network, and the 3D IDHT network includes a forward propagation algorithm of the 3D IDHT network and a backward propagation algorithm of the 3D IDHT network. Among them, the principle of the forward propagation algorithm of the 2D IDHT network, the forward propagation algorithm of the 3D IDHT network and the backward propagation algorithm of the 3D DHT network is the same, which is described in detail in the implementation manner A2, and will not be repeated here. The principle of the backward propagation algorithm of the 2D IDHT network, the backward propagation algorithm of the 3D IDHT network and the forward propagation algorithm of the 3D DHT network is the same, which is described in detail in the implementation manner A1, and will not be repeated here.

[0163] In a possible implementation manner, the first network model is adjusted based on the label image pair, the predicted image pair, the third predicted image and the fourth predicted image to obtain a second network model, including: obtaining a first loss value according to the label image pair and the predicted image pair; obtaining a second loss value according to the label image pair, the third predicted image and the fourth predicted image; obtaining a loss value of the first network model based on the first loss value and the second loss value; and adjusting the first network model based on the loss value of the first network model to obtain the second network model.

[0164] In the embodiments of the present application, the label image pair includes a first label image and a second label image, and the prediction image pair includes a first prediction image and a second prediction image. A first loss value is calculated based on the first label image, the second label image, the first prediction image and the second prediction image. A second loss value is calculated based on the first label image, the second label image, a third prediction image and a fourth prediction image, and a loss value of the first network model is calculated based on the first loss value and the second loss value. The image acquisition model is trained through the label image pair, the prediction image pair, the third prediction image and the fourth prediction image, so as to realize simultaneous supervision of model learning in the image space and the Hough space, and improve the accuracy of the model. In addition, the first loss value is calculated based on the first prediction image and the second prediction image, the second loss value is calculated based on the third prediction image and the fourth prediction image, and the first network model is adjusted through the first loss value and the second loss value, which is equivalent to supervising the model learning based on two stages of prediction images. Therefore, the model learns more information, which can accelerate the convergence speed of the model, facilitate fast model training, and improve the efficiency of image detection. The loss value of the first network model is calculated according to formula (1) as follows.

[0165]

[0166] L total =L s1 +L s2

[0167] L total is the loss value of the first network model, L s1 is the first loss value, L s2 is the second loss value, N is the number of pixel points of the heat map of the sample image in the image space, P out1 is the first prediction image, P gt is the first label image, M is the number of pixel points of the heat map of the sample image in the Hough space, H out1 is the second prediction image, H gt is the second label image, P out2 is the third prediction image, and H out2 is the fourth prediction image.

[0168] Optionally, the heat map of the sample image in the image space and the heat map of the sample image in the Hough space are obtained, the first label image is obtained by performing Gaussian processing on the heat map of the sample image in the image space, and the second label image is obtained by performing Gaussian processing on the heat map of the sample image in the Hough space.

[0169] Taking a sample image as a brain image as an example, a heat map of the brain image in an image space (the image can be a gold standard image of the brain image) is a straight line, and Gaussian processing is performed through Gaussian kernel convolution to change the straight line into a strip (that is, a bundle of lines), and a first label image is obtained after pixel values are normalized to [0, 1]. The parameters of the Gaussian kernel convolution are not limited, and for example, the parameters of the Gaussian kernel convolution are (5, 5, 5). A heat map of the brain image in a Hough space (the image is also a gold standard image of the brain image) includes a point whose position is (phi, theta, rho), and the point is converted into a Gaussian sphere through Gaussian processing, wherein the size of the Gaussian sphere is not limited, and for example, the size of the Gaussian sphere is (5, 5, 5).

[0170] The loss function shown in formula (1) is a mean square error loss function. In application, other functions except the mean square error loss function can be selected as the loss function of the first network model.

[0171] After obtaining the loss value of the first network model, the model parameters of the first network model are adjusted based on the loss value of the first network model to obtain a second network model. The second network model is taken as the image acquisition model in response to meeting a training termination condition.

[0172] The embodiment of the application provides a first network model, as shown in Figure 8 Figure 8 is a schematic diagram of a first network model provided by the embodiment of the application. In the embodiment of the application, the sample image is a brain scan image, the first label image is a heat map of the brain scan image in an image space, and the second label image is a heat map of the brain scan image in a Hough space.

[0173] The first network model can include a stem network, a hourglass network and a 3DDHT network. The brain scan image is input into the first network model, and the initial features of the brain scan image are extracted by the stem network. The stem network can include three convolution layers. Then, the initial features are processed by the hourglass network to obtain a first image space feature (that is, a first image space feature), wherein the hourglass network is a convolutional neural network used for key point detection.

[0174] ​In one aspect, the first image space feature is subjected to 1x1x1 convolution processing and up-sampling processing to obtain a heat map of the brain scan image in the image space (i.e., the first predicted image). In another aspect, the first image space feature is subjected to Hough transform processing by the 3D DHT network to obtain a first Hough space feature (i.e., the first Hough space feature). The first Hough space feature is subjected to 1x1x1 convolution processing and up-sampling processing to obtain a heat map of the brain scan image in the Hough space (i.e., the second predicted image).

[0175] Subsequently, a loss value of the first network model is obtained based on the first label image, the second label image, the first predicted image, and the second predicted image. The image acquisition model is trained based on the loss value of the first network model. Since the image acquisition model includes the 3D DHT network, the parameters of the 3D DHT network are adjusted when the first network model is trained.

[0176] Optionally, the first network model further includes a residual block network, a 3D IDHT network, etc. The first Hough space feature is first subjected to two residual block networks to further extract the features of the Hough space, and then subjected to inverse Hough transform processing by the 3D IDHT network to obtain a third image space feature.

[0177] Subsequently, the third image space feature, the first image space feature, and the initial feature are subjected to processing (i.e., fusion processing) to obtain a first fusion feature. The first fusion feature is subjected to feature processing by the hourglass network to obtain a second image space feature (i.e., the second image space feature). In one aspect, the second image space feature is subjected to 1x1x1 convolution processing and up-sampling processing to obtain a heat map of the brain scan image in the image space (i.e., the third predicted image). In another aspect, the second image space feature is subjected to Hough transform processing by the 3D DHT network to obtain a second Hough space feature (i.e., the second Hough space feature). The second Hough space feature is subjected to 1x1x1 convolution processing and up-sampling processing to obtain a heat map of the brain scan image in the Hough space (i.e., the fourth predicted image).

[0178] Subsequently, a loss value of the first network model is obtained based on the first label image, the second label image, the first to fourth predicted images according to formula (1). The image acquisition model is trained based on the loss value of the first network model. Since the image acquisition model includes the 3D DHT network and the 3D IDHT network, the parameters of the 3D DHT network and the 3D IDHT network are adjusted simultaneously in the process of training the first network model.

[0179] In one possible implementation, the first image space feature includes at least two sub-image space features; and determining the first predicted image based on the first image space feature includes determining the first predicted image based on the at least two sub-image space features.

[0180] In an embodiment of the present application, the first image space features include at least two sub-image space features, and the at least two sub-image space features are pyramid level features, that is, each sub-image space feature is progressively increased or progressively decreased, and the first network model is capable of determining the first predicted image based on the at least two sub-image space features.

[0181] Optionally, the first Hough space features are determined based on the first image space features, including: for any one of the at least two sub-image space features, determining a sub-Hough space feature corresponding to the any one sub-image space feature, and the first Hough space features include the sub-Hough space features corresponding to each sub-image space feature; and the second predicted image is determined based on the first Hough space features, including: determining the second predicted image based on the sub-Hough space features corresponding to each sub-image space feature.

[0182] For any one of the at least two sub-image space features, the Hough transform is performed on the sub-image space feature to obtain a sub-Hough space feature corresponding to the sub-image space feature. In this way, the sub-Hough space features corresponding to each sub-image space feature are obtained, that is, the first Hough space features are obtained. The first Hough space features include the sub-Hough space features corresponding to each sub-image space feature, which are also pyramid level features, and the first network model is capable of determining the second predicted image based on the sub-Hough space features corresponding to each sub-image space feature.

[0183] Optionally, the first image space features include at least two sub-image space features, and the first Hough space features include the sub-Hough space features corresponding to each sub-image space feature; the first Hough space features are determined based on the first image space features, including: in response to any one sub-image space feature being any one sub-image space feature other than the first sub-image space feature among the at least two sub-image space features, determining a second fusion feature corresponding to the any one sub-image space feature based on the any one sub-image space feature and a previous sub-image space feature of the any one sub-image space feature; and determining the sub-Hough space feature corresponding to the any one sub-image space feature based on the second fusion feature corresponding to the any one sub-image space feature and the previous sub-image space feature of the any one sub-image space feature.

[0184] In an embodiment of the present application, for the first sub-image space feature, the second fusion feature corresponding to the first sub-image space feature is determined based on the first sub-image space feature.

[0185] For any one of the at least two sub-image features except the first sub-image spatial feature, a second fusion feature corresponding to the sub-image spatial feature is determined based on the sub-image spatial feature and a previous sub-image spatial feature of the sub-image spatial feature. For example, for the third sub-image spatial feature, a second fusion feature corresponding to the third sub-image spatial feature is determined based on the third sub-image spatial feature and the second sub-image spatial feature.

[0186] Optionally, the determining of the second fusion feature corresponding to any one of the sub-image spatial features based on the any one of the sub-image spatial features and a previous sub-image spatial feature of the any one of the sub-image spatial features comprises: determining a second fusion feature corresponding to the previous sub-image spatial feature based on the previous sub-image spatial feature; determining a fourth image spatial feature based on the second fusion feature corresponding to the previous sub-image spatial feature, the fourth image spatial feature being used to represent a feature of the sample image in the image space; determining a third Hough spatial feature based on the fourth image spatial feature, the third Hough spatial feature being used to represent a feature of the sample image in the Hough space; and determining the second fusion feature corresponding to the any one of the sub-image spatial features based on the any one of the sub-image spatial features, the third Hough spatial feature and the second fusion feature corresponding to the previous sub-image spatial feature.

[0187] For any one of the at least two sub-image features except the first sub-image spatial feature, a second fusion feature corresponding to the previous sub-image spatial feature is determined. When the previous sub-image spatial feature is the first sub-image spatial feature, the second fusion feature corresponding to the first sub-image spatial feature is determined based on the first sub-image spatial feature. Optionally, the first sub-image spatial feature is convoluted to obtain the second fusion feature corresponding to the first sub-image spatial feature. When the previous sub-image spatial feature is any one of the sub-image spatial features except the first sub-image spatial feature, the second fusion feature corresponding to the previous sub-image spatial feature is determined in the same way as the second fusion feature corresponding to any one of the sub-image spatial features, which will be described in detail below.

[0188] For any one of the at least two sub-image features except the first sub-image spatial feature, after the second fusion feature corresponding to the previous sub-image spatial feature is determined, the second fusion feature corresponding to the previous sub-image spatial feature is convoluted to obtain a fourth image spatial feature. Then, the fourth image spatial feature is subjected to Hough transform to obtain a third Hough spatial feature. Subsequently, the second fusion feature corresponding to the any one of the sub-image spatial features is obtained based on the any one of the sub-image spatial features, the third Hough spatial feature and the second fusion feature corresponding to the previous sub-image spatial feature.

[0189] Optionally, the third Hough space feature is subjected to inverse Hough transform processing to obtain a fifth image space feature, and the fifth image space feature is used to represent the feature of the sample image in the image space. Then, the fifth image space feature is subjected to normalization processing to obtain a normalized fifth image space feature. After that, the normalized fifth image space feature and any sub-image space feature are multiplied to obtain a product result. The second fusion feature corresponding to the previous sub-image space feature is subjected to convolution processing to obtain a fourth image space feature. The product result and the fourth image space feature are added to obtain the second fusion feature corresponding to the sub-image space feature. For example, the normalization processing can be implemented by using an activation function such as Sigmoid.

[0190] See Figure 9 , Figure 9 is a schematic diagram of determination of the second fusion feature corresponding to any sub-image space feature provided by the embodiment of the present application. Since the second fusion feature corresponding to the previous sub-image space feature is larger than any sub-image space feature, the second fusion feature corresponding to the previous sub-image space feature is subjected to 3x3 convolution processing with a step of 2 to align the size of the second fusion feature corresponding to the previous sub-image space feature with that of any sub-image space feature, to obtain a fourth image space feature. The fourth image space feature is subjected to Hough transform processing based on a DHT network to obtain a third Hough space feature. Then, the third Hough space feature is subjected to feature extraction based on two residual block networks to further extract the features of the Hough space, and the third Hough space feature after the feature extraction is subjected to inverse Hough transform processing based on an IDHT network to obtain a fifth image space feature. Then, the fifth image space feature is subjected to 1x1 convolution processing, and the fifth image space feature after the convolution processing is subjected to normalization processing to obtain a normalized fifth image space feature.

[0191] After that, any sub-image space feature is subjected to 3x3 convolution processing to align the size of the second fusion feature corresponding to the previous sub-image space feature with that of any sub-image space feature. The convolution-processed any sub-image space feature and the normalized fifth image space feature are multiplied to obtain a product result, and the product result and the fourth image space feature are added to obtain the second fusion feature corresponding to the sub-image space feature.

[0192] After that, the second fusion feature corresponding to any sub-image space feature and the previous sub-image space feature of any sub-image space feature are used to determine a sub-Hough space feature corresponding to any sub-image space feature.

[0193] In the embodiments of the present application, the process of obtaining the normalized fifth image space feature based on the fourth image space feature can be represented by formula (2) as shown below.

[0194] y = σ (h -1 (f (h (x))) ) formula (2)

[0195] Wherein, y is the normalized fifth image space feature, sigma is the function symbol of normalization processing, h -1 is the function symbol of inverse Hough transform processing, f is the function symbol of feature extraction of two residual block networks, h is the function symbol of Hough transform processing, and x is the fourth image space feature.

[0196] It should be noted that, Figure 9 The network structure shown in the figure can be used as a pluggable network structure in the first network model. That is, the first network model can contain Figure 9 The network structure shown in the figure, or can not contain Figure 9 The network structure shown in the figure, Figure 9 The network structure shown in the figure is a network structure based on attention mechanism. The second fusion feature corresponding to the previous sub-image space feature of any sub-image space feature is a high-level feature, and any sub-image space feature is a low-level feature. By performing Hough transform processing, inverse Hough transform processing, etc. on the high-level feature, the fifth image space feature is obtained, so that the network structure pays more attention to the image space feature, improves the accuracy of the image space feature, and thus improves the accuracy of the image acquisition model.

[0197] Optionally, based on the second fusion feature corresponding to any sub-image space feature and the previous sub-image space feature of any sub-image space feature, the sub-Hough space feature corresponding to any sub-image space feature is determined, including: based on the previous sub-image space feature of any sub-image space feature, the sub-Hough space feature corresponding to the previous sub-image space feature is determined; based on the second fusion feature corresponding to any sub-image space feature, the sub-Hough space feature corresponding to the second fusion feature is determined; based on the sub-Hough space feature corresponding to the second fusion feature and the sub-Hough space feature corresponding to the previous sub-image space feature, the sub-Hough space feature corresponding to any sub-image space feature is determined.

[0198] For any one of the at least two sub-image features except the first sub-image spatial feature, a sub-Hough space feature corresponding to the previous sub-image spatial feature of the any one of the at least two sub-image features is determined. When the previous sub-image spatial feature is the first sub-image spatial feature, a second fusion feature corresponding to the first sub-image spatial feature is subjected to Hough transform processing to obtain a sub-Hough space feature corresponding to the first sub-image spatial feature. When the previous sub-image spatial feature is any one of the at least two sub-image features except the first sub-image spatial feature, the sub-Hough space feature corresponding to the previous sub-image spatial feature is determined in the same manner as the sub-Hough space feature corresponding to any one of the at least two sub-image features, and details are described below.

[0199] For any one of the at least two sub-image features except the first sub-image spatial feature, after the sub-Hough space feature corresponding to the previous sub-image spatial feature of the any one of the at least two sub-image features is determined, the second fusion feature corresponding to the any one of the at least two sub-image features is subjected to Hough transform processing to obtain a sub-Hough space feature corresponding to the second fusion feature, and then the sub-Hough space feature corresponding to the second fusion feature and the sub-Hough space feature corresponding to the previous sub-image spatial feature are fused to obtain a sub-Hough space feature corresponding to the any one of the at least two sub-image features.

[0200] Based on the above manner, the sub-Hough space feature corresponding to each sub-image spatial feature can be determined. The first image spatial feature of the embodiment of the application includes at least two sub-image spatial features, and based on the at least two sub-image spatial features, a first prediction image can be determined. The first Hough space feature includes the sub-Hough space feature corresponding to each sub-image spatial feature, and based on the sub-Hough space feature corresponding to each sub-image spatial feature, a second prediction image can be determined. Then, based on the label image pair, the first prediction image and the second prediction image, the first network model is adjusted to obtain a second network model.

[0201] Optionally, the first network model is adjusted based on the label image pair and the prediction image pair to obtain a second network model, including: obtaining a loss value of the first network model according to the label image pair and the prediction image pair; and adjusting the first network model based on the loss value of the first network model to obtain the second network model.

[0202] In the embodiment of the application, the label image pair includes a first label image and a second label image, and the prediction image pair includes a first prediction image and a second prediction image. The loss value of the first network model is calculated based on the first label image, the second label image, the first prediction image and the second prediction image, and the image acquisition model is trained through the label image pair and the prediction image pair to realize simultaneous supervision of model learning in the image space and the Hough space, which can accelerate the convergence speed of the model and improve the efficiency of image detection.

[0203] The loss value of the first network model is calculated according to formula (3) as shown below.

[0204]

[0205] L total = L spatial + L hough

[0206] L spatial is the loss value of the image space, L hough is the loss value of the Hough space, L total is the loss value of the first network model, N is the number of pixel points of the sample image in the image space, l is the first predicted image, is the first label image, M is the number of pixel points of the sample image in the Hough space, h is the second predicted image, is the second label image.

[0207] After obtaining the loss value of the first network model, the model parameters of the first network model are adjusted based on the loss value of the first network model to obtain a second network model. The second network model is taken as the image acquisition model in response to satisfying a training termination condition.

[0208] Another first network model is provided in the embodiments of the present application, as shown in Figure 10 . Figure 10 FIG. 1 is a schematic diagram of another first network model provided in the embodiments of the present application. In the embodiments of the present application, the sample image is a photographic image, the first label image is a heat map (referred to as a photographic composition) of the photographic image in the image space, and the second label image is a heat map (referred to as a Hough key point heat map) of the photographic image in the Hough space.

[0209] The first network model includes a residual network (such as a Resnet 50 network), a feature pyramid network (Feature Pyramid Networks, FPN), a Hough pyramid attention network (Hough Pyramid Attention Network, HPAN), a DHT network, and an output network. The photographic image is input to the residual network, and the initial features of the photographic image are extracted by the residual network. The initial features include at least two sub-initial features, and the at least two sub-initial features are pyramid level features.

[0210] Taking 6 sub-initial features as an example. The photographic image is input into the residual network, and the information flow of the residual network is the upstream information flow. The residual network first extracts the sub-initial feature C0, and then obtains the sub-initial feature C1 based on the sub-initial feature C0, and then obtains the sub-initial feature C2 based on the sub-initial feature C1, and so on, until the sub-initial feature C5 is obtained. Among them, the sub-initial features C2 to C5 are input into the FPN, and the first image space feature of the photographic image is determined by the FPN, and the first image space feature includes at least two sub-image space features, and the information flow of the FPN is the downstream information flow.

[0211] Taking 4 sub-image space features as an example. The FPN determines the sub-image space feature P5 based on the sub-initial feature C5, the FPN determines the sub-image space feature P4 based on the sub-initial feature C4 and the sub-image space feature P5, the FPN determines the sub-image space feature P3 based on the sub-initial feature C3 and the sub-image space feature P4, and the FPN determines the sub-image space feature P2 based on the sub-initial feature C2 and the sub-image space feature P3. On the one hand, the output network determines and outputs the photographic composition based on the sub-image space features P2 to P5, and on the other hand, the HPAN determines the second fusion feature corresponding to each sub-image space feature, and the information flow of the HPAN is the upstream information flow.

[0212] The HPAN determines the second fusion feature PA2 corresponding to the sub-image space feature P2 based on the sub-image space feature P2, the HPAN determines the second fusion feature PA3 corresponding to the sub-image space feature P3 based on the sub-image space feature P3 and the second fusion feature PA2, the HPAN determines the second fusion feature PA4 corresponding to the sub-image space feature P4 based on the sub-image space feature P4 and the second fusion feature PA3, and the HPAN determines the second fusion feature PA5 corresponding to the sub-image space feature P5 based on the sub-image space feature P5 and the second fusion feature PA4.

[0213] Among them, the HPAN is based on Figure 9The second fusion feature is determined in the illustrated manner, and taking the example of the HPAN determining the second fusion feature PA3 based on the sub-image space feature P3 and the second fusion feature PA2. The HPAN first performs 3x3 convolution processing on the second fusion feature PA2 to obtain a fourth image space feature, performs Hough transform processing on the fourth image space feature based on the DHT network to obtain a third Hough space feature. Next, the third Hough space feature is first subjected to feature extraction based on two residual block networks, and then subjected to inverse Hough transform processing based on the IDHT network to obtain a fifth image space feature. Next, the fifth image space feature is subjected to 1x1 convolution processing, and the fifth image space feature after the convolution processing is subjected to normalization processing to obtain the fifth image space feature after the normalization processing. Subsequently, the sub-image space feature P3 is subjected to 3x3 convolution processing, the sub-image space feature P3 after the convolution processing is multiplied by the fifth image space feature after the normalization processing to obtain a product result, and the product result is added to the fourth image space feature to obtain the second fusion feature PA3.

[0214] After the HPAN determines the second fusion feature corresponding to each sub-image space feature, the DHT network determines the sub-Hough space feature corresponding to each sub-image space feature, and the information flow of the DHT network is the upstream information flow.

[0215] The DHT network performs Hough transform processing on the second fusion feature PA2 to obtain the sub-Hough space feature H2 corresponding to the sub-image space feature P2. The DHT network performs Hough transform processing on the second fusion feature PA3 to obtain the sub-Hough space feature corresponding to the second fusion feature PA3, and determines the sub-Hough space feature H3 corresponding to the sub-image space feature P3 based on the sub-Hough space feature corresponding to the second fusion feature PA3 and the sub-Hough space feature H2. The DHT network performs Hough transform processing on the second fusion feature PA4 to obtain the sub-Hough space feature corresponding to the second fusion feature PA4, and determines the sub-Hough space feature H4 corresponding to the sub-image space feature P4 based on the sub-Hough space feature corresponding to the second fusion feature PA4 and the sub-Hough space feature H3. The DHT network performs Hough transform processing on the second fusion feature PA5 to obtain the sub-Hough space feature corresponding to the second fusion feature PA5, and determines the sub-Hough space feature H5 corresponding to the sub-image space feature P5 based on the sub-Hough space feature corresponding to the second fusion feature PA5 and the sub-Hough space feature H4. Subsequently, the output network determines and outputs the Hough key point heat map based on the sub-Hough space features H2 to H5.

[0216] It should be noted that when the sample image is a three-dimensional sample image, the image acquisition model mainly detects a certain plane of the three-dimensional sample image, and therefore, the heat map of the sample image in the image space can be used as the plane attention image of the sample image. When the sample image is a two-dimensional sample image, the image acquisition model mainly detects a straight line or an ellipse of the two-dimensional sample image, and therefore, the heat map of the sample image in the image space can be used as the line attention image of the sample image. Since the DHT network and the IDHT network can realize the conversion between the sample image in the image space and the sample image in the Hough space, the image acquisition model can conveniently extract geometric features such as lines and planes, and improve the accuracy of image detection.

[0217] Based on the above implementation environment, the embodiment of the present application further provides an image detection method. Figure 11 As shown in the flowchart of the image detection method provided by the embodiment of the present application, the method can be executed by the electronic device 11 in the Figure 1 As shown in the flowchart of the image detection method provided by the embodiment of the present application, the method can be executed by the electronic device 11 in the Figure 11 The method includes steps 1101 to 1103.

[0218] Step 1101, obtaining a to-be-detected image.

[0219] The embodiment of the present application does not limit the to-be-detected image, which includes but is not limited to brain images, landscape images, road images, fetal images, cell images, etc., and the number of to-be-detected images is at least one.

[0220] Step 1102, obtaining a target image of the to-be-detected image according to an image acquisition model.

[0221] The target image is a heat map of the to-be-detected image in the Hough space, and the image acquisition model is obtained according to the training method of the image acquisition model shown in the above method embodiment.

[0222] In the embodiment of the present application, the to-be-detected image is input into the image acquisition model, and the target image pair of the to-be-detected image is output by the image acquisition model. The target image pair of the to-be-detected image includes a first target image and a second target image. The first target image is a heat map of the to-be-detected image in the image space, and the value of any pixel point in the first target image represents the probability of the pixel point on the target straight line or on the target ellipse or on the target plane. The second target image is a heat map of the to-be-detected image in the Hough space, and the value of any pixel point in the second target image represents the probability of the pixel point being a target straight line or a target ellipse or a target plane. The second target image is the target image mentioned in step 1102.

[0223] It should be noted that the value of any pixel point in the first target image represents the probability of the pixel point on the target straight line (or on the target ellipse or on the target plane), if the target straight line is determined based on the first target image, a plurality of points need to be determined from the first target image, and a target straight line is formed by using the plurality of points. Since the determination of each point may have an error, the target straight line is prone to distortion, and the cumulative error of the plurality of points will make the accuracy of the target straight line lower. The value of any pixel point in the second target image represents the probability of the pixel point being the target straight line (or being the target ellipse or being the target plane), and by determining a pixel point from the second target image, a standard target straight line can be determined, and the accuracy is higher. Therefore, the method provided in the embodiments of the present application uses the second target image as the target image when detecting the image, so that the accuracy of the subsequent image detection result is higher.

[0224] In a possible implementation, the to-be-detected image is a brain image, and the target image of the to-be-detected image is obtained according to the image acquisition model, including: obtaining a first image space feature of the brain image according to the image acquisition model, the first image space feature of the brain image being used to represent the characteristics of the brain image in the image space; determining a first Hough space feature of the brain image based on the first image space feature of the brain image, the first Hough space feature of the brain image being used to represent the characteristics of the brain image in the Hough space; and determining the target image of the brain image based on the first Hough space feature of the brain image.

[0225] Optionally, the image acquisition model includes a backbone network, an hourglass network, and a 3D DHT network, the backbone network extracts an initial feature of the brain image, the first image space feature of the brain image is obtained after the initial feature of the brain image is processed by the hourglass network, the first Hough space feature of the brain image is obtained after the first image space feature of the brain image is processed by the 3D DHT network, and the target image is obtained after the first Hough space feature is processed by 1x1x1 convolution processing and up-sampling processing. For details, please refer to the related introduction of Figure 8 , which will not be repeated here.

[0226] In a possible implementation, the target image of the brain image is determined based on the first Hough space feature of the brain image, including: determining a second image space feature of the brain image based on the first Hough space feature of the brain image and the first image space feature of the brain image; determining a second Hough space feature of the brain image based on the second image space feature of the brain image; and determining the target image of the brain image based on the second Hough space feature of the brain image.

[0227] In the embodiment of the present application, after obtaining the first Hough space feature of the brain image, the image acquisition model first determines the second image space feature of the brain image based on the first image space feature of the brain image and the first Hough space feature of the brain image. Then, the second Hough space feature of the brain image is obtained by performing Hough transform processing based on the second image space feature of the brain image. Subsequently, the target image is obtained by performing up-sampling processing based on the second Hough space feature of the brain image.

[0228] Optionally, the image acquisition model comprises a backbone network, an hourglass network, a 3D DHT network, a residual block network, a 3D IDHT network, etc. After the brain image is processed by the backbone network, the hourglass network and the 3D DHT network, the first Hough space feature of the brain image is obtained. The first Hough space feature of the brain image is first processed by two residual block networks to further extract the features of the Hough space, and then is processed by the 3D IDHT network to perform inverse Hough transform processing, thereby obtaining the third image space feature of the brain image.

[0229] Subsequently, the third image space feature of the brain image, the first image space feature and the initial feature are fused to obtain the first fused feature. After the first fused feature is processed by the hourglass network, the second image space feature of the brain image is obtained. The second image space feature of the brain image is processed by the 3D DHT network to perform Hough transform processing, thereby obtaining the second Hough space feature of the brain image. After the second Hough space feature is processed by 1x1x1 convolution processing and up-sampling processing, the target image is obtained. For details, please refer to the related introduction of Figure 8 , which will not be repeated here.

[0230] It should be noted that the to-be-detected image can be a brain image, and can also be a landscape image, a road image, a fetal image, a cell image, etc. The processing mode of the image acquisition model for the landscape image, the road image, the fetal image, the cell image, etc. is similar to the processing mode of the image acquisition model for the brain image, and the processing mode of the image acquisition model for the to-be-detected image is similar to the processing mode of the first network model for the sample image. For details, please refer to the introduction of the first network model, which will not be repeated here.

[0231] In step 1103, the image detection result of the to-be-detected image is determined based on the target image.

[0232] In the embodiment of the present application, the target image is a heat map of the to-be-detected image in the Hough space. By determining the peak point in the target image, a certain plane, straight line or ellipse of the to-be-detected image is determined, thereby obtaining the image detection result.

[0233] In a possible implementation, determining the image detection result of the to-be-detected image based on the target image comprises: performing non-maximum suppression processing on the target image to obtain at least one peak point in the target image; and determining the image detection result of the to-be-detected image based on the at least one peak point in the target image.

[0234] Non-maximum suppression is to suppress elements that are not maximum values, which can be understood as local maximum value search. In the embodiment of the application, the target image is a heat map of the to-be-detected image in a Hough space, and the target image includes at least one Gaussian ball. Through non-maximum suppression processing, the maximum value in each Gaussian ball is searched to obtain a peak point of the Gaussian ball, that is, at least one peak point in the target image. Any peak point corresponds to a plane, a straight line, or an ellipse.

[0235] Optionally, the to-be-detected image is a brain image, and the image detection result is a brain midline plane of the brain image; the to-be-detected image is a fetus image, and the image detection result is at least one standard plane of the fetus image, any standard plane representing any one of a standard plane of a head of the fetus, a standard plane of an abdomen of the fetus, and a standard plane of a femur of the fetus; the to-be-detected image is a cell image, and the image detection result is at least one ellipse of the cell image, any ellipse representing a contour of a cell; and the to-be-detected image is a photographic image, and the image detection result is at least one line of the photographic image, the at least one line representing a structure of an object in the photographic image.

[0236] The to-be-detected image includes but is not limited to a brain image, a fetus image, a cell image, and a photographic image. According to different to-be-detected images, the image detection result of the to-be-detected image is also different. In the following, different to-be-detected images and their image detection results are introduced in detail from the perspective of implementation B1 to implementation B4.

[0237] In implementation B1, the to-be-detected image is a brain image, and determining the image detection result of the to-be-detected image based on the target image comprises: determining a peak point in the target image; and determining a brain midline plane based on the peak point in the target image.

[0238] In the embodiment of the application, when the to-be-detected image is a brain image, the brain image is input into an image acquisition model to obtain a target image output by the image acquisition model, where the target image is a heat map of a key point. Based on the key point in the target image, a peak point of the key point is determined to obtain a peak point in the target image. Based on the peak point in the target image, a plane equation is determined, and the plane equation is a plane equation corresponding to a brain midline plane of the brain image.

[0239] Please refer to Figure 12 , Figure 12is a processing schematic diagram of a brain scan image provided by an embodiment of the present application. Embodiments of the present application do not limit the structure of the image acquisition model. Illustratively, the image acquisition model can be a model as shown in Figure 8 , the image to be detected is a brain scan image, and the brain scan image includes at least one brain image.

[0240] The brain scan image is input to the image acquisition model, and a Hough key point heat map (i.e., a target image) is output by the image acquisition model. Peak points are calculated based on the Hough key point heat map. The peak points include three parameters, respectively θ and ρ. Among them, θ and ρ are three parameters of a plane in a polar coordinate system. According to the plane equation of the polar coordinate system , the plane equation Ax+By+Cz+D=0 of the rectangular coordinate system is determined, and the plane equation corresponding to the brain midline plane of the brain image is obtained. On the one hand, the brain midline offset is calculated based on the plane equation, and a two-dimensional visualization image is made based on the brain midline offset. The thickened line in the two-dimensional visualization image represents the brain midline. On the other hand, three-dimensional visualization is performed based on the plane equation to obtain a three-dimensional visualization image. The parallelogram in the three-dimensional visualization image represents the brain midline plane.

[0241] In implementation B2, the image to be detected is a fetal image, and an image detection result of the image to be detected is determined based on the target image, including: determining each peak point in the target image; based on each peak point in the target image, determining a standard plane corresponding to each peak point, and the standard plane corresponding to any peak point represents any one of a standard plane of a fetal head, a standard plane of a fetal abdomen, and a standard plane of a fetal femur.

[0242] In the embodiments of the present application, when the image to be detected is a fetal image, the fetal image is input to the image acquisition model to obtain a target image output by the image acquisition model, where the target image is a heat map of at least one key point. Based on each key point in the target image, a peak point of each key point is determined to obtain each peak point in the target image. Based on each peak point in the target image, a plane equation corresponding to each peak point is determined, and the plane equation corresponding to any peak point is a plane equation corresponding to a standard plane of a fetal head or a standard plane of a fetal abdomen or a standard plane of a fetal femur.

[0243] Please refer to Figure 13 , Figure 13 is a processing schematic diagram of a fetal scan image provided by an embodiment of the present application. Embodiments of the present application do not limit the structure of the image acquisition model. Illustratively, the image acquisition model can be a model as shown in Figure 8 , the image to be detected is a fetal scan image, and the fetal scan image includes at least one fetal image.

[0244] The fetal scan image is input into the image acquisition model, and a Hough key point heat map (i.e., a target image) is output by the image acquisition model, where the Hough key point heat map is a heat map of multiple key points. Non-maximum suppression processing is performed on the Hough key point heat map to search for local maximum values, so as to determine peak points of each key point, and multiple peak points are obtained. Plane equations corresponding to each peak point are determined, so as to obtain plane equations of femurs, abdomens, heads, and the like of the fetus, so as to quickly locate the plane, reduce the film reading time, and obtain standard planes of the femurs, the abdomens, the heads, and the like of the fetus. Then, fetal development parameters are measured based on the standard planes of the femurs, the abdomens, the heads, and the like of the fetus, so as to evaluate the fetal development situation.

[0245] In an implementation B3, the to-be-detected image is a cell image, and determining the image detection result of the to-be-detected image based on the target image includes: determining each peak point in the target image; and determining an ellipse corresponding to each peak point based on each peak point in the target image, where the ellipse corresponding to any peak point represents the contour of a cell.

[0246] In the embodiments of the present application, when the to-be-detected image is a cell image, the cell image is input into the image acquisition model to obtain a target image output by the image acquisition model, where the target image is a heat map of at least one key point. Based on each key point in the target image, a peak point of each key point is determined to obtain each peak point in the target image. Based on each peak point in the target image, an ellipse equation corresponding to each peak point is determined, and the ellipse equation corresponding to any peak point is an ellipse equation corresponding to the contour of a cell.

[0247] It can be understood that the cell image includes at least one cell, and by determining the ellipse equation corresponding to each peak point, the ellipse equation corresponding to the contour of each cell in the cell image can be determined, so that each cell in the cell image can be detected.

[0248] Please refer to Figure 14 , Figure 14 is a processing schematic diagram of a cell image provided by the embodiments of the present application. The embodiments of the present application do not limit the structure of the image acquisition model, and exemplarily, the image acquisition model can be a model as shown in Figure 10 , the to-be-detected image is a cell image, and the cell image includes multiple cells.

[0249] The cell image is input into the image acquisition model, and a Hough key point heat map (i.e., a target image) is output by the image acquisition model, where the Hough key point heat map is a heat map of multiple key points. Non-maximum suppression processing is performed on the Hough key point heat map to search for local maximum values, so as to determine peak points of each key point, and multiple peak points are obtained. Ellipse equations corresponding to each peak point are determined, and ellipse equations corresponding to contours of each cell in the cell image are obtained, so that each cell in the cell image is detected, and a detection result is obtained.

[0250] In an implementation B4, the image to be detected is a photographic image, and determining an image detection result of the image to be detected based on the target image includes: determining each peak point in the target image; and determining a line corresponding to each peak point based on each peak point in the target image, where the line corresponding to each peak point represents a structure of the photographic object.

[0251] In the embodiments of the present application, when the image to be detected is a photographic image, the photographic image is input into the image acquisition model, and a target image output by the image acquisition model is obtained, where the target image is a heat map of at least one key point. Based on each key point in the target image, a peak point of each key point is determined, and each peak point in the target image is obtained. Based on each peak point in the target image, a straight line equation or an ellipse equation corresponding to each peak point is determined, so that a line (a straight line or an ellipse, etc.) corresponding to each peak point is obtained. The line corresponding to any peak point is a line of the photographic object, and the lines corresponding to each peak point can represent a structure of the photographic object.

[0252] It can be understood that when the photographic image is a road image, the line corresponding to any peak point can be a line of a road edge, a line of a road marker (such as a straight road marker or a left-turn road marker), and the lines corresponding to each peak point can represent a structure of the road.

[0253] Please refer to Figure 15 , Figure 15 which is a processing schematic diagram of a photographic image provided by the embodiments of the present application. The embodiments of the present application do not limit the structure of the image acquisition model. For example, the image acquisition model can be a model as shown in Figure 10 .

[0254] The photographic image is input into the image acquisition model, and a Hough key point heat map (i.e., a target image) is output by the image acquisition model, where the Hough key point heat map is a heat map of multiple key points. Non-maximum suppression processing is performed on the Hough key point heat map to search for local maximum values, so as to determine peak points of each key point, and multiple peak points are obtained. Ellipse equations corresponding to each peak point are determined, and ellipse equations corresponding to contours of each cell in the cell image are obtained, so that each cell in the cell image is detected, and a detection result is obtained.

[0255] The image detection model in the method is trained based on a heat map of a sample image in a Hough space and a heat map of the sample image in an image space, so that the image acquisition model learns key point features in the Hough space and semantic features in the image space, can quickly and accurately determine the heat map of the image in the Hough space, accurately determine the image detection result based on the heat map of the image in the Hough space, and improve the accuracy of subsequent analysis and processing.

[0256] Figure 16 As shown in FIG. 1, the training device of the image acquisition model provided by the embodiment of the present application includes a processor 1600 and a memory 1601. Figure 16

[0257] The first acquisition module 1601 is configured to acquire a sample image label image pair, the sample image label image pair including a first label image and a second label image, the first label image being a heat map of the sample image in an image space obtained through labeling, and the second label image being a heat map of the sample image in a Hough space obtained through labeling.

[0258] The second acquisition module 1602 is configured to acquire a sample image prediction image pair according to the first network model, the sample image prediction image pair including a first prediction image and a second prediction image, the first prediction image being a heat map of the sample image in the image space obtained through prediction, and the second prediction image being a heat map of the sample image in the Hough space obtained through prediction.

[0259] The adjustment module 1603 is configured to adjust the first network model based on the sample image label image pair and the sample image prediction image pair to obtain a second network model.

[0260] The determination module 1604 is configured to, in response to a training termination condition being met, determine the second network model as an image acquisition model.

[0261] In a possible implementation, the second acquisition module 1602 is configured to acquire a first image space feature of the sample image, the first image space feature being used to represent a feature of the sample image in the image space; determine the first prediction image based on the first image space feature; determine a first Hough space feature based on the first image space feature, the first Hough space feature being used to represent a feature of the sample image in the Hough space; and determine the second prediction image based on the first Hough space feature.

[0262] In a possible implementation, the second acquisition module 1602 is configured to perform rectification processing on the first image space feature to obtain a rectified first image space feature; and determine the first Hough space feature based on the rectified first image space feature.

[0263] ​In a possible implementation, the adjusting module 1603 is configured to determine a loss value of the first network model based on the label image pair and the prediction image pair; and adjust the first network model based on the loss value of the first network model to obtain the second network model.

[0264] In a possible implementation, the second obtaining module 1602 is further configured to determine the second image space feature based on the first Hough space feature and the first image space feature; determine a third prediction image based on the second image space feature, the third prediction image being a heat map of the sample image in the image space obtained through prediction; determine a second Hough space feature based on the second image space feature; and determine a fourth prediction image based on the second Hough space feature, the fourth prediction image being a heat map of the sample image in the Hough space obtained through prediction.

[0265] The adjusting module 1603 is configured to adjust the first network model based on the label image pair, the prediction image pair, the third prediction image, and the fourth prediction image to obtain the second network model.

[0266] In a possible implementation, the second obtaining module 1602 is configured to determine the third image space feature based on the first Hough space feature; fuse the first image space feature and the third image space feature to obtain a first fused feature; and determine the second image space feature based on the first fused feature.

[0267] In a possible implementation, the adjusting module 1603 is configured to obtain a first loss value according to the label image pair and the prediction image pair; obtain a second loss value according to the label image pair, the third prediction image, and the fourth prediction image; obtain a loss value of the first network model based on the first loss value and the second loss value; and adjust the first network model based on the loss value of the first network model to obtain the second network model.

[0268] In a possible implementation, the first image space feature includes at least two sub-image space features.

[0269] The second obtaining module 1602 is configured to determine the first prediction image based on the at least two sub-image space features.

[0270] In a possible implementation, the second obtaining module 1602 is configured to, for any one of the at least two sub-image space features, determine a sub-Hough space feature corresponding to the any one of the sub-image space features, the first Hough space feature including the sub-Hough space features corresponding to the respective sub-image space features; and determine the second prediction image based on the sub-Hough space features corresponding to the respective sub-image space features.

[0271] In a possible implementation, the second acquisition module 1602 is configured to, in response to any one of the sub-image space features being any one of the sub-image space features other than the first sub-image space feature, determine a second fusion feature corresponding to the any one of the sub-image space features based on the any one of the sub-image space features and a previous sub-image space feature of the any one of the sub-image space features; and determine a sub-Hough space feature corresponding to the any one of the sub-image space features based on the second fusion feature corresponding to the any one of the sub-image space features and the previous sub-image space feature.

[0272] In a possible implementation, the second acquisition module 1602 is configured to determine a second fusion feature corresponding to a previous sub-image space feature based on the previous sub-image space feature; determine a fourth image space feature based on the second fusion feature corresponding to the previous sub-image space feature; determine a third Hough space feature based on the fourth image space feature; and determine a second fusion feature corresponding to any one of the sub-image space features based on the any one of the sub-image space features, the third Hough space feature, and the second fusion feature corresponding to the previous sub-image space feature.

[0273] In a possible implementation, the second acquisition module 1602 is configured to determine a sub-Hough space feature corresponding to a previous sub-image space feature based on the previous sub-image space feature; determine a sub-Hough space feature corresponding to a second fusion feature based on the second fusion feature; and determine a sub-Hough space feature corresponding to any one of the sub-image space features based on the sub-Hough space feature corresponding to the second fusion feature and the sub-Hough space feature corresponding to the previous sub-image space feature.

[0274] The device image detection model is trained based on the heat map of the sample image in the Hough space and the heat map of the sample image in the image space, so that the image acquisition model learns the key point features in the Hough space and the semantic features in the image space, can quickly and accurately determine the heat map of the image in the Hough space, accurately determine the image detection result based on the heat map of the image in the Hough space, and improve the accuracy of subsequent analysis and processing.

[0275] It should be understood that the above Figure 16 When the apparatuses provided in the embodiments are implemented, the above-described functions can be allocated to different functional modules in the apparatuses, that is, the internal structure of the apparatuses is divided into different functional modules to complete all or part of the above-described functions. In addition, the apparatuses provided in the embodiments and the method embodiments belong to the same concept, and the specific implementation process is described in the method embodiments, which will not be described here.

[0276] Figure 17 Fig. 1 shows a structural schematic diagram of an image detection device provided by an embodiment of the present application, as shown in the figure, the device comprises: Figure 17

[0277] A first obtaining module 1701 is configured to obtain a to-be-detected image.

[0278] A second obtaining module 1702 is configured to obtain a target image of the to-be-detected image according to an image obtaining model, the target image being a heat map of the to-be-detected image in a Hough space, the image obtaining model being obtained according to the training method of any of the above image obtaining models.

[0279] A determining module 1703 is configured to determine an image detection result of the to-be-detected image based on the target image.

[0280] In a possible implementation, the to-be-detected image is a brain image, and the second obtaining module 1702 is configured to obtain a first image space feature of the brain image according to the image obtaining model, the first image space feature of the brain image being used to represent a feature of the brain image in an image space; determine a first Hough space feature of the brain image based on the first image space feature of the brain image, the first Hough space feature of the brain image being used to represent a feature of the brain image in a Hough space; and determine the target image of the brain image based on the first Hough space feature of the brain image.

[0281] In a possible implementation, the second obtaining module 1702 is configured to determine a second image space feature of the brain image based on the first Hough space feature of the brain image and the first image space feature of the brain image; determine a second Hough space feature of the brain image based on the second image space feature of the brain image; and determine the target image of the brain image based on the second Hough space feature of the brain image.

[0282] In a possible implementation, the determining module 1703 is configured to perform a non-maximum suppression process on the target image to obtain at least one peak point in the target image; and determine the image detection result of the to-be-detected image based on the at least one peak point in the target image.

[0283] The above device image detection model is trained based on a heat map of a sample image in a Hough space and a heat map of the sample image in an image space, so that the image obtaining model learns key point features in the Hough space and semantic features in the image space, can quickly and accurately determine a heat map of an image in the Hough space, accurately determine an image detection result based on the heat map of the image in the Hough space, and thus improve the accuracy of subsequent analysis and processing.

[0284] It should be understood that the above Figure 17 ​The apparatus provided in the embodiments of the present application is only used as an example to illustrate the division of the functional modules, and in actual applications, the above-mentioned functions can be completed by different functional modules according to the needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0285] Figure 18 A structure block diagram of a terminal device 1800 provided by an example embodiment of the present application is shown. The terminal device 1800 can be a portable mobile terminal, such as a smartphone, a tablet computer, an MP3 (Moving Picture Experts Group Audio Layer III) player, an MP4 (Moving Picture Experts Group Audio Layer IV) player, a notebook computer, or a desktop computer. The terminal device 1800 can also be referred to as a user equipment, a portable terminal, a laptop terminal, a desktop terminal, or other names.

[0286] Generally, the terminal device 1800 includes a processor 1801 and a memory 1802.

[0287] The processor 1801 can include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 1801 can be implemented in at least one of a hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), and a PLA (Programmable Logic Array). The processor 1801 can also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also known as a CPU (Central Processing Unit). The coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 1801 can be integrated with a GPU (Graphics Processing Unit) that is responsible for rendering and drawing the content to be displayed by the display screen. In some embodiments, the processor 1801 can also include an AI (Artificial Intelligence) processor for processing machine learning-related computing operations.

[0288] The memory 1802 can include one or more computer-readable storage media. The computer-readable storage media can be non-transitory. The memory 1802 can also include high-speed random access memory and can include nonvolatile memory, such as one or more magnetic disk storage devices, optical storage devices, flash memory devices, or other nonvolatile solid-state storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 1802 is used to store at least one instruction for execution by the processor 1801 to implement the training method of the image acquisition model or the image detection method provided by the method embodiments of the present application.

[0289] In some embodiments, the terminal device 1800 can also optionally include a peripheral device interface 1803 and at least one peripheral device. The processor 1801, the memory 1802, and the peripheral device interface 1803 can be connected through a bus or a signal line. Each peripheral device can be connected to the peripheral device interface 1803 through a bus, a signal line, or a circuit board. Specifically, the peripheral device includes at least one of a radio frequency circuit 1804, a display screen 1805, a camera assembly 1806, an audio circuit 1807, a positioning assembly 1808, and a power supply 1809.

[0290] The peripheral device interface 1803 can be used to connect at least one peripheral device related to input / output (I / O) to the processor 1801 and the memory 1802. In some embodiments, the processor 1801, the memory 1802, and the peripheral device interface 1803 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1801, the memory 1802, and the peripheral device interface 1803 can be implemented on a separate chip or circuit board, and the present embodiment is not limited in this regard.

[0291] The radio frequency circuit 1804 is configured to receive and send RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 1804 communicates with communication networks and other communication devices through electromagnetic signals. The radio frequency circuit 1804 converts electrical signals to electromagnetic signals for transmission, or vice versa. Optionally, the radio frequency circuit 1804 includes an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a subscriber identity module card, and the like. The radio frequency circuit 1804 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes, but is not limited to, the World Wide Web, a metropolitan area network, an intranet, various generations of mobile communication networks (2G, 3G, 4G, and 5G), a wireless local area network, and / or a WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 1804 can also include NFC (Near Field Communication) related circuitry, which is not limited in the present application.

[0292] The display screen 1805 is configured to display a UI (User Interface). The UI can include graphics, text, icons, video, and any combination thereof. When the display screen 1805 is a touch display screen, the display screen 1805 also has the ability to collect touch signals on or above the surface of the display screen 1805. The touch signals can be input as control signals to the processor 1801 for processing. At this time, the display screen 1805 can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or soft keyboards. In some embodiments, the display screen 1805 can be one, arranged on the front panel of the terminal device 1800; in other embodiments, the display screen 1805 can be at least two, arranged on different surfaces of the terminal device 1800 or in a folding design; in other embodiments, the display screen 1805 can be a flexible display screen, arranged on a curved surface or a folding surface of the terminal device 1800. Even, the display screen 1805 can also be arranged in an irregular shape, that is, a special-shaped screen. The display screen 1805 can be made of LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), and the like.

[0293] The camera component 1806 is configured to capture images or videos. Optionally, the camera component 1806 includes a front-facing camera and a rear-facing camera. Generally, the front-facing camera is disposed on the front panel of the terminal, and the rear-facing camera is disposed on the back of the terminal. In some embodiments, the rear-facing camera is at least two, which is any one of a main camera, a depth-of-field camera, a wide-angle camera, and a telephoto camera, to realize the background blur function of the main camera and the depth-of-field camera, the panorama shooting and VR (Virtual Reality) shooting function of the main camera and the wide-angle camera, or other fusion shooting functions. In some embodiments, the camera component 1806 can also include a flash. The flash can be a single-color-temperature flash or a dual-color-temperature flash. The dual-color-temperature flash refers to the combination of a warm light flash and a cold light flash, which can be used for light compensation at different color temperatures.

[0294] The audio circuit 1807 can include a microphone and a speaker. The microphone is configured to capture sound waves of a user and an environment, and convert the sound waves into an electrical signal input to the processor 1801 for processing or to the radio frequency circuit 1804 to realize voice communication. For the purpose of stereo sound collection or noise reduction, the microphone can be multiple, which are respectively disposed at different parts of the terminal device 1800. The microphone can also be an array microphone or an omnidirectional collection microphone. The speaker is configured to convert an electrical signal from the processor 1801 or the radio frequency circuit 1804 into sound waves. The speaker can be a traditional diaphragm speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert an electrical signal into a sound wave audible to humans, but also convert an electrical signal into an inaudible sound wave to humans for ranging purposes. In some embodiments, the audio circuit 1807 can also include a headphone jack.

[0295] The positioning component 1808 is configured to locate the current geographical position of the terminal device 1800 to realize navigation or LBS (Location Based Service).

[0296] The power supply 1809 is configured to supply power to each component in the terminal device 1800. The power supply 1809 can be alternating current, direct current, a disposable battery, or a rechargeable battery. When the power supply 1809 includes a rechargeable battery, the rechargeable battery can be a wired charging battery or a wireless charging battery. The wired charging battery is a battery charged through a wired line, and the wireless charging battery is a battery charged through a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0297] In some embodiments, the terminal device 1800 further includes one or more sensors 1810. The one or more sensors 1810 include, but are not limited to, an acceleration sensor 1811, a gyroscope sensor 1812, a pressure sensor 1813, a fingerprint sensor 1814, an optical sensor 1815, and a proximity sensor 1816.

[0298] The acceleration sensor 1811 can detect the acceleration magnitude in three coordinate axes of a coordinate system established by the terminal device 1800. For example, the acceleration sensor 1811 can be used to detect the components of the gravitational acceleration in three coordinate axes. The processor 1801 can control the display screen 1805 to display the user interface in a landscape view or a portrait view according to the gravitational acceleration signals collected by the acceleration sensor 1811. The acceleration sensor 1811 can also be used for game or user motion data collection.

[0299] The gyroscope sensor 1812 can detect the body orientation and rotation angle of the terminal device 1800. The gyroscope sensor 1812 can work with the acceleration sensor 1811 to collect the 3D motion of the user to the terminal device 1800. The processor 1801 can implement the following functions according to the data collected by the gyroscope sensor 1812: motion sensing (e.g., changing the UI according to the user's tilt operation), image stabilization when shooting, game control, and inertial navigation.

[0300] The pressure sensor 1813 can be arranged on the side frame of the terminal device 1800 and / or the lower layer of the display screen 1805. When the pressure sensor 1813 is arranged on the side frame of the terminal device 1800, the user's holding signal to the terminal device 1800 can be detected, and the left-hand or right-hand recognition or shortcut operation can be performed by the processor 1801 according to the holding signal collected by the pressure sensor 1813. When the pressure sensor 1813 is arranged on the lower layer of the display screen 1805, the processor 1801 can control the operable control on the UI interface according to the user's pressure operation to the display screen 1805. The operable control includes at least one of a button control, a scroll bar control, an icon control, and a menu control.

[0301] The fingerprint sensor 1814 is configured to collect a fingerprint of a user, and the processor 1801 identifies an identity of the user according to the fingerprint collected by the fingerprint sensor 1814, or the fingerprint sensor 1814 identifies the identity of the user according to the collected fingerprint. When the identity of the user is identified as a trusted identity, the processor 1801 authorizes the user to perform a related sensitive operation, which includes unlocking a screen, viewing encrypted information, downloading software, payment, and changing settings, etc. The fingerprint sensor 1814 can be arranged on the front, back or side of the terminal device 1800. When a physical button or a manufacturer's logo is arranged on the terminal device 1800, the fingerprint sensor 1814 can be integrated with the physical button or the manufacturer's logo.

[0302] The optical sensor 1815 is configured to collect an ambient light intensity. In an embodiment, the processor 1801 can control the display brightness of the display screen 1805 according to the ambient light intensity collected by the optical sensor 1815. Specifically, when the ambient light intensity is high, the display brightness of the display screen 1805 is increased; when the ambient light intensity is low, the display brightness of the display screen 1805 is decreased. In another embodiment, the processor 1801 can also dynamically adjust the shooting parameters of the camera assembly 1806 according to the ambient light intensity collected by the optical sensor 1815.

[0303] The proximity sensor 1816, also known as a distance sensor, is usually arranged on the front panel of the terminal device 1800. The proximity sensor 1816 is configured to collect the distance between the user and the front of the terminal device 1800. In an embodiment, when the proximity sensor 1816 detects that the distance between the user and the front of the terminal device 1800 gradually decreases, the processor 1801 controls the display screen 1805 to switch from a bright screen state to an off-screen state; when the proximity sensor 1816 detects that the distance between the user and the front of the terminal device 1800 gradually increases, the processor 1801 controls the display screen 1805 to switch from the off-screen state to the bright screen state.

[0304] Those skilled in the art can understand that the structure shown in the above description is not a limitation on the terminal device 1800, and the terminal device 1800 can include more or fewer components than those shown in the figure, or combine certain components, or use different component arrangements. Figure 18

[0305] Figure 19 ​A structural diagram of a server provided in the embodiments of the present application is shown in FIG. 19. The server 1900 can have great differences due to different configurations or performances, and can include one or more processors 1901 and one or more memories 1902. The one or more memories 1902 store at least one piece of program code, which is loaded and executed by the one or more processors 1901 to implement the image acquisition model training method or the image detection method provided in any of the above-mentioned embodiments. The processor 1901 is, for example, a CPU. Of course, the server 1900 can also have a wired or wireless network interface, a keyboard, an input and output interface, and other components for implementing device functions, which are not described herein.

[0306] In an exemplary embodiment, a computer readable storage medium is also provided, which stores at least one piece of program code. The at least one piece of program code is loaded and executed by a processor to enable an electronic device to implement any of the above-mentioned image acquisition model training methods or image detection methods.

[0307] Optionally, the computer readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, or the like.

[0308] In an exemplary embodiment, a computer program or computer program product is also provided, which stores at least one computer instruction. The at least one computer instruction is loaded and executed by a processor to enable a computer to implement any of the above-mentioned image acquisition model training methods or image detection methods.

[0309] It should be understood that "multiple" referred to herein means two or more. The "and / or" describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after it.

[0310] The serial numbers of the above embodiments of the present application are only for description, and do not represent the advantages or disadvantages of the embodiments.

[0311] The above merely describes exemplary embodiments of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for training an image acquisition model, characterized in that, The method comprises: obtaining a label image pair of a sample image, the label image pair comprising a first label image and a second label image, the first label image being a heat map of the sample image in an image space obtained through labeling, and the second label image being a heat map of the sample image in a Hough space obtained through labeling; obtaining a first image space feature of the sample image according to a first network model, the first image space feature being used to represent a feature of the image space; determining a first predicted image in a predicted image pair based on the first image space feature; determining a first Hough space feature based on the first image space feature, the first Hough space feature being used to represent a feature of the Hough space; determining a second predicted image in the predicted image pair based on the first Hough space feature, the first predicted image being a heat map of the sample image in the image space obtained through prediction, and the second predicted image being a heat map of the sample image in the Hough space obtained through prediction; determining a second image space feature based on the first Hough space feature and the first image space feature; determining a third predicted image which is a heat map of the sample image in the image space obtained through prediction based on the second image space feature; determining a second Hough space feature based on the second image space feature; determining a fourth predicted image which is a heat map of the sample image in the Hough space obtained through prediction based on the second Hough space feature; adjusting the first network model based on the label image pair, the predicted image pair, the third predicted image and the fourth predicted image to obtain a second network model; in response to satisfying a training termination condition, taking the second network model as an image acquisition model; wherein the process of determining the first predicted image, the second predicted image, the third predicted image and the fourth predicted image comprises up-sampling processing.

2. The method of claim 1, wherein, The process of determining the first Hough space feature based on the first image space feature comprises: rectifying the first image space feature to obtain a rectified first image space feature; determining the first Hough space feature based on the rectified first image space feature.

3. The method of claim 1, wherein, The process of determining the second image space feature based on the first Hough space feature and the first image space feature comprises: determining a third image space feature based on the first Hough space feature; fusing the first image space feature and the third image space feature to obtain a first fused feature; determining the second image space feature based on the first fused feature.

4. The method of claim 1, wherein, The process of adjusting the first network model based on the label image pair, the predicted image pair, the third predicted image and the fourth predicted image to obtain a second network model comprises: obtaining a first loss value according to the label image pair and the predicted image pair; obtaining a second loss value according to the label image pair, the third predicted image and the fourth predicted image; obtaining a loss value of the first network model based on the first loss value and the second loss value; adjust the first network model based on the loss value of the first network model to obtain a second network model.

5. The method according to claim 1 or 2, characterized in that, The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features.

6. The method of claim 5, wherein, The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features.

7. The method according to claim 1 or 2, characterized in that, The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features.

8. The method of claim 7, wherein, The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features.

9. The method of claim 7, wherein, The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub-image space features. The first image space feature includes at least two sub determine the sub-Hough space feature corresponding to the any sub-image space feature based on the sub-Hough space feature corresponding to the second fusion feature and the sub-Hough space feature corresponding to the previous sub-image space feature.

10. An image detection method characterized by, The method comprises: obtaining a to-be-detected image; obtaining a target image of the to-be-detected image according to an image obtaining model, the target image being a heat map of the to-be-detected image in a Hough space, the image obtaining model being obtained according to the method in any one of claims 1 to 9; determining an image detection result of the to-be-detected image based on the target image.

11. The method of claim 10, wherein, The to-be-detected image is a brain image, and the obtaining of the target image of the to-be-detected image according to the image obtaining model comprises: obtaining a first image space feature of the brain image according to the image obtaining model, the first image space feature of the brain image being used to represent the feature of the brain image in the image space; determining a first Hough space feature of the brain image based on the first image space feature of the brain image, the first Hough space feature of the brain image being used to represent the feature of the brain image in the Hough space; determining the target image of the brain image based on the first Hough space feature of the brain image.

12. The method of claim 11, wherein, The determining of the target image of the brain image based on the first Hough space feature of the brain image comprises: determining a second image space feature of the brain image based on the first Hough space feature of the brain image and the first image space feature of the brain image; determining a second Hough space feature of the brain image based on the second image space feature of the brain image; determining the target image of the brain image based on the second Hough space feature of the brain image.

13. The method of claim 10, wherein, The determining of the image detection result of the to-be-detected image based on the target image comprises: performing non-maximum suppression processing on the target image to obtain at least one peak point in the target image; determining the image detection result of the to-be-detected image based on the at least one peak point in the target image.

14. A training apparatus for an image acquisition model, characterized in that, The device comprises: a first obtaining module, configured to obtain a label image pair of a sample image, the label image pair comprising a first label image and a second label image, the first label image being a heat map of the sample image in an image space obtained through labeling, and the second label image being a heat map of the sample image in a Hough space obtained through labeling; a second obtaining module, configured to obtain a first image space feature of the sample image according to a first network model, the first image space feature being used to represent the feature of the image space; determine a first predicted image in a predicted image pair based on the first image space feature; determine a first Hough space feature based on the first image space feature, the first Hough space feature being used to represent the feature of the Hough space; and determine a second predicted image in the predicted image pair based on the first Hough space feature, the first predicted image being a heat map of the sample image in the image space obtained through prediction, and the second predicted image being a heat map of the sample image in the Hough space obtained through prediction. The second acquisition module is further configured to determine a second image space feature based on the first Hough space feature and the first image space feature; determine a third predicted image based on the second image space feature, the third predicted image being a heat map of the sample image in image space obtained by prediction; determine a second Hough space feature based on the second image space feature; and determine a fourth predicted image based on the second Hough space feature, the fourth predicted image being a heat map of the sample image in Hough space obtained by prediction. The adjusting module is configured to adjust the first network model based on the pair of label images, the pair of predicted images, the third predicted image, and the fourth predicted image to obtain a second network model. The determining module is configured to, in response to a training termination condition being met, determine the second network model as an image acquisition model. The process of determining the first predicted image, the second predicted image, the third predicted image, and the fourth predicted image includes up-sampling processing.

15. An image detection apparatus characterized by comprising: The apparatus comprises: The first acquisition module is configured to acquire a to-be-detected image. The second acquisition module is configured to acquire a target image of the to-be-detected image according to an image acquisition model, the target image being a heat map of the to-be-detected image in Hough space, the image acquisition model being obtained according to the method of any one of claims 1 to 9. The determining module is configured to determine an image detection result of the to-be-detected image based on the target image.

16. An electronic device, comprising: The electronic device comprises a processor and a memory, the memory storing at least one program code, the at least one program code being loaded and executed by the processor, so that the electronic device implements the image acquisition model training method of any one of claims 1 to 9 or implements the image detection method of any one of claims 10 to 13.

17. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one program code, the at least one program code being loaded and executed by the processor, so that the computer implements the image acquisition model training method of any one of claims 1 to 9 or implements the image detection method of any one of claims 10 to 13.

18. A computer program product, characterized in that the computer program product stores at least one computer instruction, the at least one computer instruction being loaded and executed by a processor to implement the image acquisition model training method of any one of claims 1 to 9 or implement the image detection method of any one of claims 10 to 13.

Images