Adaptive image classification method, device, equipment and storage medium

Abstract

The application relates to the field of image recognition classification, and discloses a self-adaptive image classification method, device, equipment and storage medium, the method comprising the following steps: acquiring a current medical image; performing self-adaptive classification processing on the current medical image through a preset deep neural network model to obtain a classification processing result; and determining the type of the current medical image according to the classification processing result. The preset deep neural network model is used to perform self-adaptive classification processing on the image, so that the image type corresponding to the image is obtained, and compared with the prior art, the simple image or the complex image can be effectively distinguished, so that the effect of improving the image classification precision and reducing the classification time is achieved.

Description

Technical Field

[0001] This invention relates to the field of image recognition and classification, and in particular to an adaptive image classification method, apparatus, device, and storage medium. Background Technology

[0002] With the rapid development of deep learning neural networks in image recognition technology, deep learning-based intelligent image recognition has become widely used in various image recognition application scenarios. By realizing intelligent recognition of medical case images, deep learning has been promoted and applied in the field of medical case recognition. In existing technologies, image data is often processed using relatively traditional methods. Such methods are usually difficult to process large amounts of image data efficiently, and the accuracy of image classification is not satisfactory.

[0003] Therefore, improving image classification accuracy and reducing image classification time has become an urgent problem to be solved.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] This invention provides an adaptive image classification method, apparatus, device, and storage medium, aiming to solve the technical problems of low image classification accuracy and long image classification time in the prior art.

[0006] To achieve the above objectives, the present invention provides an adaptive image classification method, the method comprising the following steps:

[0007] Acquire current medical images;

[0008] The current medical image is classified using an adaptive preset deep neural network model to obtain the classification result; the network layer in the preset deep neural network model includes 5 sub-layers; the sub-layer includes 2 3×3 convolutional layers and 1 2×2 pooling layer, and the pooling layer is after the 2 convolutional layers;

[0009] Based on the classification results, the type of the current medical image is determined.

[0010] Optionally, the step of classifying the current medical image using an adaptive preset deep neural network model to obtain the classification result includes:

[0011] When the preset deep neural network model acquires the current medical image, the first feedback depth of the first sub-layer in the preset deep neural network model is adaptively selected, and the next feedback depth of the first feedback depth is obtained.

[0012] The feature entropy value of the current medical image is calculated based on the first feedback depth and the next feedback depth, respectively, to obtain the processed first image feature entropy value and second image feature entropy value;

[0013] The second image feature entropy value is compared with the first image feature entropy value to determine whether the first image feature entropy value has decreased.

[0014] If so, the feedback depth corresponding to the first image feature entropy value is selected as the output depth of the first sub-layer, and the same processing method as the first sub-layer is performed through subsequent sub-layers.

[0015] Once the output depth of all sub-layers has been selected, the current medical image is classified according to the output depth of the corresponding sub-layer to obtain the classification result.

[0016] Optionally, after the step of comparing the second image feature entropy value with the first image feature entropy value to determine whether the first image feature entropy value has decreased, the method further includes:

[0017] If not, then obtain the training feedback depth and the first feedback depth when training the deep neural network model;

[0018] The first feedback depth is compared with the training feedback depth. When the next feedback depth of the first feedback depth is less than the training feedback depth, the feature entropy value calculated by the next feedback depth of the first feedback depth is compared with the first image feature entropy value to determine whether the first image feature entropy value has decreased.

[0019] If so, the feedback depth corresponding to the first image feature entropy value is selected as the output depth of the first sub-layer, and the same processing as the first sub-layer is performed through subsequent sub-layers.

[0020] Optionally, after the step of obtaining the training feedback depth and the first feedback depth when training the deep neural network model (if not), the method further includes:

[0021] Determine whether the next feedback depth of the first feedback depth is greater than or equal to the training feedback depth;

[0022] If the next feedback depth is greater than or equal to the training feedback depth, then the training feedback depth is directly selected as the output depth of the first sub-layer, and the same processing method as the first sub-layer is performed through subsequent sub-layers.

[0023] Optionally, before acquiring the initial medical image, the process further includes:

[0024] Obtain the initial medical image dataset;

[0025] The medical images in the initial medical image dataset are preprocessed to obtain the preprocessed target medical images;

[0026] The target medical image is used to iteratively train a deep neural network model to obtain a preset deep neural network model.

[0027] Optionally, the step of preprocessing the medical images in the initial medical image dataset to obtain the preprocessed target medical image includes:

[0028] The medical images in the initial medical image dataset are randomly cropped to obtain cropped images after cropping.

[0029] The cropped image is randomly horizontally flipped to obtain the flipped target medical image.

[0030] Optionally, the deep neural network model further includes a loss function, which is a cross-entropy loss function. The cross-entropy loss function is used to calculate the training parameter error and update the parameter weights so that the model has the optimal weight result when it reaches the number of iterations.

[0031] Furthermore, to achieve the above objectives, the present invention also proposes an adaptive image classification device, the device comprising:

[0032] The image acquisition module is used to acquire the current medical image;

[0033] The identification and classification module is used to adaptively classify the current medical image through a preset deep neural network model to obtain the classification result; the network layer in the preset deep neural network model includes 5 sub-layers; the sub-layer includes 2 3×3 convolutional layers and 1 2×2 pooling layer, and the pooling layer is after the 2 convolutional layers;

[0034] The type acquisition module is used to determine the type of the current medical image based on the classification processing result.

[0035] Furthermore, to achieve the above objectives, the present invention also proposes an adaptive image classification device, the device comprising: a memory, a processor, and an adaptive image classification program stored in the memory and executable on the processor, the adaptive image classification program being configured to implement the steps of the adaptive image classification method as described above.

[0036] Furthermore, to achieve the above objectives, the present invention also proposes a storage medium storing an adaptive image classification program, which, when executed by a processor, implements the steps of the adaptive image classification method as described above.

[0037] This invention acquires a current medical image; performs adaptive classification processing on the current medical image using a preset deep neural network model to obtain a classification result; the preset deep neural network model includes five sub-layers; each sub-layer includes two 3×3 convolutional layers and one 2×2 pooling layer, with the pooling layer following the two convolutional layers; based on the classification result, the type of the current medical image is determined. This invention uses a preset deep neural network model to adaptively classify images to obtain the corresponding image type. Compared to existing technologies, it can effectively distinguish between simple and complex images, thereby improving image classification accuracy and reducing classification time. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the structure of an adaptive image classification device with an adaptive hardware operating environment involved in the embodiments of the present invention;

[0039] Figure 2 This is a flowchart illustrating a first embodiment of an adaptive image classification method according to the present invention.

[0040] Figure 3 This is a schematic diagram of the adaptive deep neural network model structure in the first embodiment of the adaptive image classification method of the present invention;

[0041] Figure 4 This is a flowchart illustrating a second embodiment of an adaptive image classification method according to the present invention;

[0042] Figure 5 This is a schematic diagram of the adaptive selection process in a second embodiment of an adaptive image classification method of the present invention;

[0043] Figure 6 This is a flowchart illustrating a third embodiment of an adaptive image classification method according to the present invention;

[0044] Figure 7 This is a flowchart illustrating the fourth embodiment of an adaptive image classification method of the present invention;

[0045] Figure 8 This is a block diagram of the structure of an adaptive image classification device according to the present invention.

[0046] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0047] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0048] Reference Figure 1 , Figure 1 This is a schematic diagram of the adaptive image classification device structure for the hardware operating environment involved in the embodiments of the present invention.

[0049] like Figure 1 As shown, the adaptive image classification device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk storage device. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0050] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on adaptive image classification devices and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0051] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and an adaptive image classification program.

[0052] exist Figure 1 In the adaptive image classification device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the adaptive image classification device of the present invention can be set in the adaptive image classification device, and the adaptive image classification device calls the adaptive image classification program stored in the memory 1005 through the processor 1001 and executes the adaptive image classification method provided in the embodiment of the present invention.

[0053] This invention provides an adaptive image classification method, referring to... Figure 2 , Figure 2This is a flowchart illustrating the first embodiment of an adaptive image classification method according to the present invention.

[0054] In this embodiment, the adaptive image classification method includes the following steps:

[0055] Step S100: Acquire the current medical image;

[0056] It should be noted that the executing entity of the method in this embodiment can be an electronic device with data processing, network communication, and program execution functions, such as a tablet or computer, or other electronic devices capable of performing the same or similar functions. This embodiment does not impose any limitations on this. Here, we will use an image recognition and classification device as an example to describe various embodiments of the adaptive image classification method of the present invention.

[0057] It is understood that the current medical image may be one or more medical images captured by the user at the current moment.

[0058] Step S200: The current medical image is classified using an adaptive preset deep neural network model to obtain the classification result; the network layer in the preset deep neural network model includes 5 sub-layers; the sub-layer includes 2 3×3 convolutional layers and 1 2×2 pooling layer, and the pooling layer is after the 2 convolutional layers;

[0059] It should be understood that the above-mentioned adaptation means that in the process of image processing and analysis, the neural network model can automatically adjust the processing method, processing order, processing parameters, boundary conditions or constraints according to the data characteristics of the data, so as to achieve the best processing effect on the processed data. The adaptive process of data processing is a computational process that infinitely approximates the target data. This process is represented by a mathematical model to generate an adaptive algorithm. The adaptive algorithm can be designed according to the set optimal criteria. The corresponding optimal criteria can be the minimum mean square error criterion (LMS algorithm), the least squares criterion (LS algorithm), the maximum signal-to-noise ratio criterion, and the statistical detection criterion.

[0060] It should be noted that the aforementioned preset deep neural network model can use a 13-layer convolutional neural network. The first 10 layers of the 13-layer convolutional neural network are convolutional layers, and the last 3 layers are fully connected layers. The convolutional layers perform cross-correlation operations on the matrix of each channel from left to right (the convolutional kernel is generally a 3x3 matrix) and from top to bottom through the convolution kernel (first from left to right, then from top to bottom, so the convolution operation also preserves positional information). The cross-correlation operation is to multiply and then add the corresponding positions, and finally add the values ​​of the three channels together to get a single value. The fully connected layers are responsible for transforming the two-dimensional feature map output by the convolution into a one-dimensional vector, that is, performing calculations on the previous convolution, pooling, and other layers to obtain the feature space mapping sample label space, thereby reducing the influence of feature position on the classification result and improving the robustness of the entire network.

[0061] It should be noted that, referring to Figure 3 , Figure 3 This is a schematic diagram of the adaptive deep neural network model structure in the first embodiment of the adaptive image classification method of the present invention.

[0062] The left-hand structure diagram in the figure shows the image input features, Maxpool represents max pooling, FC represents a fully connected layer, and FC-4096 indicates that this fully connected layer has 4096 perceptrons. Softmax indicates that this layer uses softpooling to pool the image features, and it is based on the softmax weighting method to preserve the basic attributes of the input while amplifying stronger feature activations. The specific parameters of the Conv3-channel on the right side of the figure are shown in the left-hand structure diagram (Conv3-64, Conv3-128, Conv3-256, Conv3-512). The Conv3-channel can be called a DFRNN structure and can be represented by the following formula:

[0063]

[0064] In the formula, a i The coefficient is represented by a. i =Sigomd(h i f(·) is the tanh() activation function, and U and W are trainable shared weights generated by the convolution kernel.

[0065] It should be noted that the aforementioned perceptron is a neuron in a neural network, and a fully connected neural network is formed by combining neurons.

[0066] Furthermore, the max pooling method described above can downsample the image, reduce the dimensionality of the image to remove redundant information, and compress features to simplify network complexity.

[0067] It should also be noted that, Figure 3 In the left-hand structure diagram, convolutional layers 2, 4, 6, 8, and 10 are weight-sharing layers. Weight-sharing layers can reduce the number of parameters and the amount of computation.

[0068] In practical implementation, when the image recognition and classification device acquires a medical image, it can input the image into a preset deep neural network model for classification processing. The preset deep neural network model can adaptively identify image features based on the parameters trained on the model, so that the image recognition device can determine the type of tea disease based on the recognition results.

[0069] Step S300: Determine the type of the current medical image based on the classification processing result.

[0070] In practice, based on the recognition and classification results, the image recognition and classification device can classify the features of the current medical image and output the medical pathology type after feature classification.

[0071] This embodiment acquires a current medical image; then performs adaptive classification processing on the current medical image using a preset deep neural network model to obtain a classification result; the network layer in the preset deep neural network model includes 5 sub-layers; each sub-layer includes two 3×3 convolutional layers and one 2×2 pooling layer, with the pooling layer following the two convolutional layers; based on the classification result, the type of the current medical image is determined. This invention uses a preset deep neural network model to adaptively classify images to obtain the corresponding image type. Compared to existing technologies, it can effectively distinguish between simple and complex images, thereby improving image classification accuracy and reducing classification time.

[0072] refer to Figure 4 , Figure 4 This is a flowchart illustrating a second embodiment of an adaptive image classification method according to the present invention.

[0073] Based on the first embodiment described above, in this embodiment, to perform adaptive recognition processing on the current medical image, step S200 further includes:

[0074] Step S210: When the preset deep neural network model obtains the current medical image, the first feedback depth of the first sub-layer in the preset deep neural network model is adaptively selected, and the next feedback depth of the first feedback depth is obtained.

[0075] It should be understood that the first sub-layer mentioned above is the first sub-layer of the preset deep neural network model. The first sub-layer has a first and a second convolutional layer, and the preset deep neural network model can adaptively select the feedback depth value as the convolution depth of the second convolutional layer according to the features of the current medical image.

[0076] It should be noted that the first feedback depth mentioned above is the convolution depth of the second convolutional layer. The larger the value of the feedback depth, the longer the time for image classification processing will be. When using a deep neural network model to classify images, the feedback depth can be adaptively selected for complex and simple images, thereby reducing the image classification time to a certain extent.

[0077] Step S220: Calculate the feature entropy value of the current medical image based on the first feedback depth and the next feedback depth respectively, to obtain the processed first image feature entropy value and second image feature entropy value;

[0078] It should be understood that the aforementioned feature entropy value is a statistical form of image features, which can reflect the average amount of information in image features.

[0079] Step S230: Compare the second image feature entropy value with the first image feature entropy value to determine whether the first image feature entropy value has decreased;

[0080] It should be understood that when the aforementioned feature entropy value increases, it means that the image processed by the previous feedback depth has a greater average information content of image features. Furthermore, by selecting the lowest possible feedback depth for image classification while retaining more average information content of image features, the classification time can be reduced.

[0081] Step S240: If yes, then select the feedback depth corresponding to the first image feature entropy value as the output depth of the first sub-layer, and continue to perform the same processing method as the first sub-layer through subsequent sub-layers;

[0082] It should be understood that the above-mentioned process of continuing to perform the same processing as the first sub-layer through subsequent sub-layers indicates that there are 5 sub-layers in the deep neural network model. After the input image features are adaptively selected in the first sub-layer to obtain the output depth of the corresponding second convolutional layer in the first sub-layer, the second, third, fourth, and fifth sub-layers of the deep neural network model will sequentially adaptively select the feedback depth of the corresponding sub-layer for the image features and use the feedback depth as the output depth of the sub-layer.

[0083] Step S250: When the output depth of all sub-layers has been selected, the current medical image is classified according to the output depth of the corresponding sub-layer to obtain the classification result.

[0084] Understandably, once all sub-layers have selected their corresponding output depths, the deep neural network model can perform convolution processing on the current image based on the corresponding output depths, and then classify the current image based on the image features obtained from the convolution processing, thereby obtaining the classification result.

[0085] In this embodiment, when the current medical image is obtained by the preset deep neural network model, the first feedback depth of the first sub-layer in the preset deep neural network model is adaptively selected, and the next feedback depth of the first feedback depth is obtained. The feature entropy value of the current medical image is calculated according to the first feedback depth and the next feedback depth to obtain the processed first image feature entropy value and second image feature entropy value. The second image feature entropy value is compared with the first image feature entropy value to determine whether the first image feature entropy value has decreased. If so, the feedback depth corresponding to the first image feature entropy value is selected as the output depth of the first sub-layer, and the same processing method as the first sub-layer is continued through subsequent sub-layers. When the output depth of all sub-layers is selected, the current medical image is classified according to the output depth of the corresponding sub-layer to obtain the classification processing result. This realizes that when using a deep neural network model to identify and classify image features, the feedback depth during image feature processing is adaptively selected, thereby effectively improving classification accuracy and reducing classification time.

[0086] refer to Figure 6 , Figure 6 This is a flowchart illustrating a third embodiment of an adaptive image classification method of the present invention.

[0087] Based on the second embodiment described above, in this embodiment, in order to reduce the computational load of the preset deep neural network model and thus reduce the image classification time, after step S230, the following is also included:

[0088] Step S231: If not, obtain the training feedback depth and the first feedback depth when training the deep neural network model;

[0089] It should be understood that the above-mentioned training feedback depth is a preset feedback depth for the deep neural network model when the image is being trained, and the deep neural network model can perform convolution processing on the image features according to the preset feedback depth.

[0090] It should be noted that the training feedback depth is used when classifying the current image. When the deep neural network model adaptively selects the feedback depth, it may exceed the training feedback depth. When this happens, it will cause the image classification process to take longer. Therefore, the training feedback depth is selected as the preset feedback depth threshold of the deep neural network model, thereby reducing the image recognition and classification time.

[0091] In a specific implementation, after step S231, the following is also included:

[0092] Step S2311: Determine whether the next feedback depth of the first feedback depth is greater than or equal to the training feedback depth;

[0093] Step S2312: If the next feedback depth is greater than or equal to the training feedback depth, the training feedback depth is directly selected as the output depth of the first sub-layer, and the same processing method as the first sub-layer is continued through subsequent sub-layers.

[0094] It should be noted that if the next feedback depth is less than the training feedback depth, the feature entropy value of the next feedback depth is assigned to the first feedback depth, and the first feedback depth is compared with the training feedback depth. When the next feedback depth of the first feedback depth is less than the training feedback depth, the feature entropy value calculated by the next feedback depth of the first feedback depth is compared with the first image feature entropy value to determine whether the first image feature entropy value has decreased. If not, the training feedback depth and the first feedback depth are obtained when training the deep neural network model, until the loop is exited. For example, if the training depth is 5, then the image should have a feature entropy value (1,2,3,4,5) every time it passes through the depth (1,2,3,4,5) of the DFRNN structure. However, our network compares in real time; it is not that the image needs to pass through depth 5 and then select a depth for output. Instead, at the beginning, it goes through depths 1 and 2 to obtain entropy values ​​1 and 2, and then compares them. If entropy value 2 is greater than entropy value 1, the depth corresponding to entropy value 1 is selected for output, the judgment ends, and the judgment proceeds to the next sub-layer judgment. If entropy value 2 is less than entropy value 1, it continues to compare entropy value 3 with entropy value 2. If 3 is greater than 2, the depth of entropy value 2 is output, and the judgment ends; otherwise, it continues to compare entropy value 4 with 3. If 4 is greater than 3, the depth of entropy value 3 is output, and the judgment ends; otherwise, it continues to compare entropy value 5 with 4. If 5 is greater than 4, the depth of 4 is output, and the judgment ends; otherwise, depth 5 is directly output.

[0095] Step S232: Compare the first feedback depth with the training feedback depth. When the next feedback depth of the first feedback depth is less than the training feedback depth, compare the feature entropy value calculated by the next feedback depth of the first feedback depth with the first image feature entropy value to determine whether the first image feature entropy value has decreased.

[0096] Step S233: If yes, then select the feedback depth corresponding to the first image feature entropy value as the output depth of the first sub-layer, and continue to perform the same processing as the first sub-layer through subsequent sub-layers.

[0097] This embodiment compares the second image feature entropy value with the first image feature entropy value to determine whether the first image feature entropy value has decreased. If the first feature map entropy value remains unchanged, it obtains the training feedback depth and the first feedback depth during training of the deep neural network model. It compares the first feedback depth with the training feedback depth; if the next feedback depth of the first feedback depth is less than the training feedback depth, it compares the feature entropy value calculated at the next feedback depth with the first image feature entropy value to determine whether the first image feature entropy value has decreased. When the first image feature entropy value decreases, it selects the next feedback depth as the output depth of the first sub-layer and continues to perform the same processing as the first sub-layer through subsequent sub-layers. This reduces the computational load of the preset deep neural network model, thereby reducing image classification time.

[0098] refer to Figure 7 , Figure 7 This is a flowchart illustrating the fourth embodiment of an adaptive image classification method of the present invention.

[0099] Based on the first embodiment described above, in this embodiment, in order to perform preprocessing of the medical image set and increase the number of samples, before step S100, the method further includes:

[0100] Step S10: Obtain the initial medical image dataset;

[0101] Understandably, the aforementioned initial medical image dataset is a collection of medical images collected by the user for training deep neural network models.

[0102] Step S20: Preprocess the medical images in the initial medical image dataset to obtain the preprocessed target medical image;

[0103] It should be noted that the preprocessing may include random flipping, random cropping, random angle rotation, color jittering, and adding noise.

[0104] It is understandable that the aforementioned target medical images are all images that have been preprocessed using image recognition and classification equipment.

[0105] Further, step S20 specifically includes:

[0106] Step S21: Randomly crop the medical images in the initial medical image dataset to obtain cropped images after cropping.

[0107] Step S22: Randomly flip the cropped image horizontally to obtain the flipped target medical image.

[0108] It should be noted that the above-mentioned random horizontal flipping process flips the image horizontally according to probability. In addition, the image can also be flipped vertically according to probability. After horizontal or vertical flipping, the relative position of the lesions does not change. This method can simulate images taken by a camera at different angles. Random cropping involves randomly selecting different regions of the image to increase data diversity. Besides using the flipping and cropping methods to obtain pre-processed images, random angle rotation, color jittering, and noise addition can also be used for image preprocessing. Random angle rotation involves rotating the image at a certain angle to obtain images from different angles to enrich the sample data. Color jittering modifies four attributes of the image: brightness, contrast, saturation, and color. Since hue is an important characteristic for distinguishing tea lesions, the image brightness is modified to simulate images taken under different lighting conditions. Noise addition involves randomly adding Gaussian noise to the image to obtain images with different levels of sharpness, simulating different shooting qualities. The above preprocessing can significantly increase the amount of image sample data. For example, if the initial sample data is 2175 images, the amount of image data obtained after preprocessing can reach more than 10,000 images.

[0109] Step S30: Iteratively train the target medical image on the deep neural network model to obtain a preset deep neural network model.

[0110] Understandably, by iteratively training a deep neural network model using target medical images, the model can gradually increase the weights for image recognition based on image features, thereby enabling the model to achieve a certain level of recognition accuracy and reducing errors during image recognition.

[0111] This embodiment acquires an initial medical image dataset; randomly crops the medical images in the initial medical image dataset to obtain a cropped image; and randomly horizontally flips the cropped image to obtain a flipped target medical image. This expands a small set of image samples into a larger set of image samples.

[0112] refer to Figure 8 , Figure 8 This is a block diagram of the structure of an adaptive image classification device according to the present invention.

[0113] Image acquisition module 801 is used to acquire the current medical image;

[0114] The identification and classification module 802 is used to adaptively classify the current medical image through a preset deep neural network model to obtain the classification result; the network layer in the preset deep neural network model includes 5 sub-layers; the sub-layer includes 2 3×3 convolutional layers and 1 2×2 pooling layer, and the pooling layer is after the 2 convolutional layers;

[0115] The type acquisition module 803 is used to determine the type of the current medical image based on the classification processing result.

[0116] This embodiment acquires a current medical image; then performs adaptive classification processing on the current medical image using a preset deep neural network model to obtain a classification result; the network layer in the preset deep neural network model includes 5 sub-layers; each sub-layer includes two 3×3 convolutional layers and one 2×2 pooling layer, with the pooling layer following the two convolutional layers; based on the classification result, the type of the current medical image is determined. This invention uses a preset deep neural network model to adaptively classify images to obtain the corresponding image type. Compared to existing technologies, it can effectively distinguish between simple and complex images, thereby improving image classification accuracy and reducing classification time.

[0117] Various embodiments or specific implementations of the adaptive image classification device of the present invention can be referred to the above-described method embodiments, and will not be repeated here.

[0118] It should be noted that, in this document, the terms "comprising," "including," or any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0119] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0120] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention can essentially be said to contribute to the prior art in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0121] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An adaptive image classification method, characterized in that, The method includes the following steps: Acquire current medical images; The current medical image is adaptively classified using an adaptive preset deep neural network model to obtain the classification result; the network layer in the preset deep neural network model includes 5 sub-layers; the sub-layer includes 2 3×3 convolutional layers and 1 2×2 pooling layer, and the pooling layer is after the 2 convolutional layers; Based on the classification results, the type of the current medical image is determined; The step of adaptively classifying the current medical image using an adaptive preset deep neural network model to obtain the classification result includes: When the preset deep neural network model acquires the current medical image, the first feedback depth of the first sub-layer in the preset deep neural network model is adaptively selected, and the next feedback depth of the first feedback depth is obtained. The feature entropy value of the current medical image is calculated based on the first feedback depth and the next feedback depth, respectively, to obtain the processed first image feature entropy value and second image feature entropy value; The second image feature entropy value is compared with the first image feature entropy value to determine whether the first image feature entropy value has decreased. If so, the feedback depth corresponding to the first image feature entropy value is selected as the output depth of the first sub-layer, and the same processing method as the first sub-layer is performed through subsequent sub-layers. Once the output depth of all sub-layers has been selected, the current medical image is classified according to the output depth of the corresponding sub-layer to obtain the classification result.

2. The method as described in claim 1, characterized in that, After the step of comparing the second image feature entropy value with the first image feature entropy value to determine whether the first image feature entropy value has decreased, the method further includes: If not, then obtain the training feedback depth and the first feedback depth when training the deep neural network model; The first feedback depth is compared with the training feedback depth. When the next feedback depth of the first feedback depth is less than the training feedback depth, the feature entropy value calculated by the next feedback depth of the first feedback depth is compared with the first image feature entropy value to determine whether the first image feature entropy value has decreased. If so, the feedback depth corresponding to the first image feature entropy value is selected as the output depth of the first sub-layer, and the same processing as the first sub-layer is performed through subsequent sub-layers.

3. The method as described in claim 2, characterized in that, If not, then after the step of obtaining the training feedback depth and the first feedback depth when training the deep neural network model, the method further includes: Determine whether the next feedback depth of the first feedback depth is greater than or equal to the training feedback depth; If the next feedback depth is greater than or equal to the training feedback depth, then the training feedback depth is directly selected as the output depth of the first sub-layer, and the same processing method as the first sub-layer is performed through subsequent sub-layers.

4. The method as described in claim 1, characterized in that, Before acquiring the current medical image, the process also includes: Obtain the initial medical image dataset; The medical images in the initial medical image dataset are preprocessed to obtain the preprocessed target medical images; The target medical image is used to iteratively train a deep neural network model to obtain a preset deep neural network model.

5. The method as described in claim 4, characterized in that, The step of preprocessing the medical images in the initial medical image dataset to obtain the preprocessed target medical image includes: The medical images in the initial medical image dataset are randomly cropped to obtain cropped images after cropping. The cropped image is randomly horizontally flipped to obtain the flipped target medical image.

6. The method as described in claim 4, characterized in that, The deep neural network model also includes a loss function, which is a cross-entropy loss function. The cross-entropy loss function is used to calculate the training parameter error and update the parameter weights so that the model has the optimal weight result when it reaches the number of iterations.

7. An adaptive image classification device, characterized in that, The device includes: The image acquisition module is used to acquire the current medical image; The identification and classification module is used to adaptively classify the current medical image through a preset deep neural network model to obtain the classification result; the network layer in the preset deep neural network model includes 5 sub-layers; the sub-layer includes 2 3×3 convolutional layers and 1 2×2 pooling layer, and the pooling layer is after the 2 convolutional layers; The type acquisition module is used to determine the type of the current medical image based on the classification processing result; The identification and classification module is further configured to: adaptively select a first feedback depth of a first sub-layer in the preset deep neural network model when the preset deep neural network model acquires the current medical image; obtain a next feedback depth of the first feedback depth; calculate the feature entropy value of the current medical image based on the first feedback depth and the next feedback depth respectively to obtain a processed first image feature entropy value and a second image feature entropy value; compare the second image feature entropy value with the first image feature entropy value to determine whether the first image feature entropy value has decreased; if so, select the feedback depth corresponding to the first image feature entropy value as the output depth of the first sub-layer, and continue to perform the same processing method as the first sub-layer through subsequent sub-layers; when the output depth of all sub-layers has been selected, perform feature classification on the current medical image based on the output depth corresponding to the respective sub-layer to obtain a classification processing result.

8. An adaptive image classification device, characterized in that, The device includes: a memory, a processor, and an adaptive image classification program stored in the memory and executable on the processor, the adaptive image classification program being configured to implement the steps of the adaptive image classification method as claimed in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores an adaptive image classification program, which, when executed by a processor, implements the steps of the adaptive image classification method as described in any one of claims 1 to 6.

Images