Engine detection method, device, medium and system

Abstract

The present application provides an engine detection method, device, medium and system. The method controls the shooting device to take multiple engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and obtains the engine image, thereby ensuring that the subsequent running trajectory of the shooting device can completely capture all the appearance surfaces of the engine, and performs data collection, and uses the image recognition model to predict the engine image, and obtain multiple resolution images of the target part and the corresponding position confidence and type confidence. According to the position confidence and type confidence of the target part, it is determined whether the engine is abnormal, thereby realizing an efficient target detection algorithm that takes into account lightweight, reasoning speed and accuracy, and realizing efficient detection of the consistency of the engine appearance, thereby solving the problem that the existing solution needs to add manual auxiliary detection when detecting whether the parts outside the engine are complete, resulting in poor detection efficiency.

Description

Technical Field

[0001] The present application relates to the technical field of engine detection, and in particular, to an engine detection method, device, medium and system. Background Art

[0002] Before the engine product is assembled and packaged for sale, it is necessary to conduct a final inspection of the engine appearance to prevent the mis-installation or missing of external parts, thereby improving product consistency and quality. The number and method of assembling external engine parts with different order numbers on the same production line are different, which makes manual inspection time-consuming and labor-intensive and prone to misjudgment.

[0003] That is, the existing solution requires manual auxiliary inspection when detecting whether the external parts of the engine are complete, which leads to poor detection efficiency. Summary of the invention

[0004] The main purpose of the present application is to provide an engine detection method, device, medium and system, so as to at least solve the problem that the existing solutions require manual auxiliary detection when detecting whether the external parts of the engine are complete, resulting in poor detection efficiency.

[0005] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a method for detecting an engine is provided, the method comprising: controlling a shooting device to take a plurality of engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and acquiring the engine image; using an image recognition model to predict the engine image, and obtain a plurality of resolution images of a target part and corresponding position confidence and type confidence; determining whether the engine is abnormal according to the position confidence and the type confidence of the target part, the engine abnormality indicating that the target part does not exist on the engine, or the target part on the engine is in an incorrect position.

[0006] Optionally, the image recognition model includes a first DLA-ELAN structure, a first MPConv structure, a second DLA-ELAN structure, a second MPConv structure, a third DLA-ELAN structure and a CSPNet structure. In the process of using the image recognition model to predict the engine image to obtain multiple resolution images of the target part and the corresponding position confidence and type confidence, the method also includes: performing three convolution processes on the engine image in sequence to obtain a first convolution result; using the first DLA-ELAN structure to process the first convolution result to obtain a first processing result; using the first MPConv structure and the second DLA-ELAN structure in sequence to process the first processing result to obtain a second processing result; using the second MPConv structure, the third DLA-ELAN structure and the CSPNet structure in sequence to process the second processing result to obtain a third processing result.

[0007] Optionally, the image recognition model includes a fourth DLA-ELAN structure. In the process of using the image recognition model to predict the engine image and obtain multiple resolution images of the target part and the corresponding position confidence and type confidence, the process includes: after obtaining the third processing result, convolving the first processing result to obtain a second convolution result, and convolving the second processing result to obtain a third convolution result; sequentially convolving and upsampling the third processing result to obtain a fourth processing result, wherein the upsampling represents an image magnification process; merging the fourth processing result and the third convolution result to obtain a first merged result; using the fourth DLA-ELAN structure to process the first merged result to obtain a fifth processing result; sequentially convolving and upsampling the fifth processing result to obtain a sixth processing result; merging the second convolution result and the sixth processing result to obtain a second merged result.

[0008] Optionally, the image recognition model includes a fifth DLA-ELAN structure, a third MPConv structure, a sixth DLA-ELAN structure, a fourth MPConv structure, a seventh DLA-ELAN structure, a first RepConv structure, a second RepConv structure and a third RepConv structure, and the image recognition model is used to predict the engine image to obtain multiple resolution images of the target part and corresponding position confidences and type confidences, including: after obtaining the second merged result, sequentially using the fifth DLA-ELAN structure and the third MPConv structure to process the second merged result to obtain a seventh processing result; merging the fifth processing result and the seventh processing result to obtain a third merged result;

[0009] The sixth DLA-ELAN structure and the fourth MPConv structure are used in sequence to process the third merged result to obtain an eighth processing result; the second MPConv structure and the third DLA-ELAN structure are used in sequence to process the second processing result to obtain a ninth processing result, and the eighth processing result and the ninth processing result are merged to obtain a fourth merged result; the seventh DLA-ELAN structure and the first RepConv structure are used in sequence to process the fourth merged result to obtain a first resolution image and corresponding position confidence and type confidence, and the sixth DLA-ELAN structure and the second RepConv structure are used in sequence to process the third merged result to obtain a second resolution image and corresponding position confidence and type confidence, and the fifth DLA-ELAN structure and the third RepConv structure are used in sequence to process the second merged result to obtain a third resolution image and corresponding position confidence and type confidence.

[0010] Optionally, the first convolution result is processed using a first DLA-ELAN structure to obtain a first processing result, including: performing convolution processing on the first convolution result to obtain a first intermediate convolution result, and performing segmentation processing on the first intermediate convolution result to obtain a first segmentation result and a second segmentation result; processing the second segmentation result using the first DLA structure to obtain a first intermediate processing result; performing convolution processing on the first intermediate processing result to obtain a second intermediate convolution result, and processing the second intermediate convolution result using the second DLA structure to obtain a second intermediate processing result; performing convolution processing on the second intermediate processing result to obtain a third intermediate convolution result; and merging the first intermediate convolution result, the first segmentation result, the second intermediate convolution result and the second intermediate processing result to obtain the first processing result.

[0011] Optionally, determining whether the engine is abnormal is based on the position confidence and the type confidence of the target part, including: when the type corresponding to the maximum value of all the type confidences in the target part is the target type, determining that the type of the target part is the target type; when the position confidence of the target part is greater than or equal to a confidence threshold, determining that the engine is normal; when the position confidence of the target part is less than a confidence threshold, determining that the engine is abnormal.

[0012] Optionally, the shooting device is controlled to shoot multiple engine images around the engine with the bearing of the flywheel housing of the engine as the axis, including: obtaining an initial flywheel housing image of the engine; using Hough circle detection technology and morphological threshold to determine the position in the initial flywheel housing image to obtain the current axis position; when the current axis position is the center of the initial flywheel housing image, the shooting device is controlled to shoot multiple engine images around the engine with the bearing of the flywheel housing of the engine as the axis; when the current axis position is not the center of the initial flywheel housing image, the position of the shooting device is adjusted until the current axis position is the center of the initial flywheel housing image.

[0013] According to another aspect of the present application, an engine detection device is provided, which includes: a first processing unit, used to control a shooting device to shoot multiple engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and obtain the engine image; a second processing unit, used to use an image recognition model to predict the engine image, and obtain multiple resolution images of the target part and the corresponding position confidence and type confidence; a third processing unit, used to determine whether the engine is abnormal based on the position confidence and the type confidence of the target part, and the engine abnormality indicates that the target part does not exist on the engine, or the target part on the engine is in the wrong position.

[0014] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the methods described.

[0015] According to another aspect of the present application, an engine detection system is provided, the system comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include methods for executing any one of the described methods.

[0016] By applying the technical solution of the present application, a shooting device is controlled to take a plurality of engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and the engine image is obtained, thereby ensuring that the subsequent operation trajectory of the shooting device can completely capture all the appearance surfaces of the engine, and data collection is performed, and an image recognition model is used to predict the engine image to obtain multiple resolution images of the target part and the corresponding position confidence and type confidence. According to the position confidence and type confidence of the target part, it is determined whether the engine is abnormal, thereby realizing an efficient target detection algorithm that takes into account lightweight, reasoning speed and accuracy, and realizing efficient detection of the consistency of the engine appearance, thereby solving the problem that the existing solution needs to add manual auxiliary detection when detecting whether the parts outside the engine are complete, resulting in poor detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings constituting part of the present application are used to provide a further understanding of the present application. The exemplary embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 A schematic flow chart of an engine detection method provided according to an embodiment of the present application is shown;

[0019] Figure 2 A schematic diagram showing the working principle of an image recognition model provided according to an embodiment of the present application;

[0020] Figure 3 A schematic diagram showing the working principle of the DLA-ELAN structure provided according to an embodiment of the present application is shown;

[0021] Figure 4 A schematic diagram of a process of model training provided according to an embodiment of the present application is shown;

[0022] Figure 5 A schematic flow chart of another engine detection method provided according to an embodiment of the present application is shown;

[0023] Figure 6 A structural block diagram of an engine detection device provided according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0025] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] For the convenience of description, some nouns or terms involved in the embodiments of the present application are explained below:

[0028] Hough circle detection technology is an image processing technology used to detect circular objects in images. This technology uses Hough transform to detect the circular contour in the image and determine the circle's position, radius and other parameters.

[0029] Morphological thresholding is an image processing method used to segment an image into different regions or extract objects of interest. This method is based on the morphological features of the image, such as shape, size, etc., and sets a threshold to segment or extract the image. Morphological thresholding can be used in applications such as image enhancement, object detection, and edge detection.

[0030] A CCD camera is a camera that uses a charge-coupled device (CCD) as an image sensor. CCD is a semiconductor device used to capture light signals and convert them into electrical signals. It is commonly used in digital cameras, camcorders, and other imaging devices. CCD cameras have the advantages of high sensitivity, low noise, and high dynamic range. They are widely used in scientific research, medical imaging, astronomy, and other fields.

[0031] GAN is the abbreviation of Generative Adversarial Network. It is a deep learning model consisting of two parts: the generator and the discriminator. The generator is responsible for generating data, while the discriminator is responsible for judging the difference between the generated data and the real data. Through continuous adversarial learning, the competition between the generator and the discriminator prompts the generator to continuously improve the quality of the generated data. GAN has achieved good results in tasks such as image generation, style transfer, and super-resolution.

[0032] The DLA structure (Deep Layer Aggregation) is a deep learning network structure for image segmentation and target detection, which improves model performance through multi-scale feature fusion and dense connections.

[0033] The MPConv structure (Multi-Path Convolutional Networks) is a multi-path convolutional network structure that introduces multiple branches and paths into the network to increase the width and depth of the network, thereby improving the representation ability and performance of the model.

[0034] The CSPNet structure (Cross Stage Partial Networks) is a network structure used for target detection. By introducing partial connections and cross-stage information transmission at different stages of the network, feature reuse and information exchange are achieved, thereby improving the accuracy and efficiency of the model.

[0035] Upsampling is a data processing technique used to handle datasets with imbalanced classes. In feature upsampling, the number of samples of each class in the dataset is balanced by increasing the number of samples of the minority class. This can be achieved by duplicating existing minority class samples or synthesizing new minority class samples. Feature upsampling helps improve the model's ability to identify the minority class and improves the model's performance and generalization ability.

[0036] As introduced in the background technology, before the engine product is assembled with all parts and ready for packaging and sale, it is necessary to conduct a final inspection of the engine appearance to prevent the external parts from being misassembled or missing, thereby improving the consistency and quality of the product. The number and method of assembling external engine parts with different order numbers on the same production line are different, which makes manual inspection time-consuming and labor-intensive and prone to misjudgment. In order to solve the problem that the existing solution needs to add manual auxiliary inspection when detecting whether the external parts of the engine are complete, resulting in poor detection efficiency, the embodiments of the present application provide an engine detection method, device, medium and system.

[0037] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0038] In this embodiment, a method for detecting an engine is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0039] Figure 1 FIG. 1 is a flow chart of an engine detection method provided according to an embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0040] Step S101, controlling the photographing device to take a plurality of engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and acquiring the engine images;

[0041] In step S101, the control shooting device takes the bearing of the flywheel housing of the engine as the axis and shoots a plurality of engine images around the engine, including:

[0042] Acquire an initial flywheel housing image of the engine;

[0043] The Hough circle detection technology and morphological threshold are used to determine the position in the initial flywheel housing image to obtain the current axis position;

[0044] When the current axis position is the center of the initial flywheel housing image, controlling the photographing device to take a plurality of images of the engine around the engine with the bearing of the flywheel housing of the engine as the axis;

[0045] When the current axis position is not the center of the initial flywheel housing image, the position of the photographing device is adjusted until the current axis position is the center of the initial flywheel housing image.

[0046] Specifically, the shooting device can be a CCD camera. By adjusting the position of the shooting device, the current axis position can be made the center of the above-mentioned initial flywheel housing image, thereby ensuring that the subsequently shot images are easier to identify, thereby ensuring that the subsequent running trajectory of the mechanical arm carrying the CCD camera can completely capture all the exterior surfaces of the engine and collect data.

[0047] The objects to be tested include: bolts, studs, nameplates, steel stamp numbers, starters, belts, heat shields, and assembly quantities of air vents.

[0048] Until a sufficient amount of image data is obtained. For targets with a small number of categories, GAN (Generative Adversarial Network) will be used to imitate the original data to generate virtual data, thereby achieving the functions of sample enhancement and category balance. Finally, the entire data set is divided into training set and test set in a ratio of 7:3.

[0049] Step S102, using an image recognition model to predict the above engine image, to obtain multiple resolution images of the target part and corresponding position confidence and type confidence;

[0050] Specifically, since the surface of the engine product has been painted and is smooth, the grayscale value of the detection point is almost the same as the background, and the shape is complex, and there are even cases where the features are partially blocked. As a result, the scene of the present invention cannot be processed using the traditional visual image processing method mentioned in the above patent. In addition, the pallet on which the engine is placed also has certain processing errors and the placement angle error of the pallet on the plate chain line. This will cause the contours of engines with the same order number to be different in the image.

[0051] The target detection algorithm based on deep learning has scale invariance and translation continuity, which can effectively overcome the changes in the position and size of the target for accurate identification. In addition, the deep learning algorithm can automatically extract deep texture, color, shape and other features in the image, and has better target recognition capabilities under complex background conditions such as the outer surface of the engine.

[0052] Construct a target detection algorithm model based on deep learning. The principle is as follows Figure 2 As shown, the sixth MPConv structure and the eighth DLA-ELAN structure are the same as the second MPConv structure and the third DLA-ELAN structure, and are not described in detail here.

[0053] In one embodiment of the present application, the image recognition model includes a first DLA-ELAN structure, a first MPConv structure, a second DLA-ELAN structure, a second MPConv structure, a third DLA-ELAN structure and a CSPNet structure. In step S102, in which the image recognition model is used to predict the engine image to obtain multiple resolution images of the target part and corresponding position confidences and type confidences, the method further includes:

[0054] The engine image is subjected to three convolution operations in sequence to obtain a first convolution result;

[0055] Using the first DLA-ELAN structure to process the first convolution result, to obtain a first processing result;

[0056] sequentially using the first MPConv structure and the second DLA-ELAN structure to process the first processing result to obtain a second processing result;

[0057] The second processing result is processed in sequence using the second MPConv structure, the third DLA-ELAN structure and the CSPNet structure to obtain a third processing result.

[0058] Specifically, the CSPNet structure is used to alleviate the problems of complex mapping and gradient confusion in tandem deep learning models.

[0059] In one embodiment of the present application, the image recognition model includes a fourth DLA-ELAN structure. In step S102, in which the image recognition model is used to predict the engine image to obtain multiple resolution images of the target part and corresponding position confidences and type confidences, the method further includes:

[0060] After obtaining the third processing result, performing convolution processing on the first processing result to obtain a second convolution result, and performing convolution processing on the second processing result to obtain a third convolution result;

[0061] The third processing result is sequentially subjected to convolution processing and up-sampling processing to obtain a fourth processing result, wherein the up-sampling processing represents an image enlargement processing;

[0062] Combining the fourth processing result and the third convolution result to obtain a first combined result;

[0063] Using the fourth DLA-ELAN structure to process the first merging result, a fifth processing result is obtained;

[0064] Performing convolution processing and upsampling processing on the fifth processing result in sequence to obtain a sixth processing result;

[0065] The second convolution result and the sixth processing result are combined to obtain a second combined result.

[0066] By adding convolution, upsampling, and the fourth DLA-ELAN structure, the three feature maps output by the subsequent three RepConv structures are more in line with actual working conditions.

[0067] In one embodiment of the present application, the image recognition model includes a fifth DLA-ELAN structure, a third MPConv structure, a sixth DLA-ELAN structure, a fourth MPConv structure, a seventh DLA-ELAN structure, a first RepConv structure, a second RepConv structure, and a third RepConv structure. Step S102, i.e., using the image recognition model to predict the engine image, obtains multiple resolution images of the target part and corresponding position confidences and type confidences, including:

[0068] After obtaining the second merging result, sequentially using the fifth DLA-ELAN structure and the third MPConv structure to process the second merging result to obtain a seventh processing result;

[0069] Combining the fifth processing result and the seventh processing result to obtain a third combined result;

[0070] sequentially using the sixth DLA-ELAN structure and the fourth MPConv structure to process the third merging result to obtain an eighth processing result;

[0071] sequentially using the second MPConv structure and the third DLA-ELAN structure to process the second processing result to obtain a ninth processing result, and merging the eighth processing result and the ninth processing result to obtain a fourth merging result;

[0072] The seventh DLA-ELAN structure and the first RepConv structure are used in sequence to process the fourth merged result to obtain a first resolution image and a corresponding position confidence and a type confidence. At the same time, the sixth DLA-ELAN structure and the second RepConv structure are used in sequence to process the third merged result to obtain a second resolution image and a corresponding position confidence and a type confidence. At the same time, the fifth DLA-ELAN structure and the third RepConv structure are used in sequence to process the second merged result to obtain a third resolution image and a corresponding position confidence and a type confidence.

[0073] Specifically, the three RepConv structures achieve the decoupling of the model training parameters and the model reasoning parameters, reducing the model reasoning time without affecting the accuracy. Finally, the first RepConv structure will output a feature map with a structure of 20×20×18, the second RepConv structure will output a feature map with a structure of 40×40×18, and the third RepConv structure will output a feature map with a structure of 80×80×18. The final output resolutions are 20×20, 40×40, and 80×80. Each feature point corresponds to the recognition result of a detection point, where the number of channels of each feature point is (5+4)×2, where 5 represents the center point coordinates, width, height, and target confidence of the prediction box, 4 represents the probability of the four categories of the prediction box, and 2 represents regression reasoning by two anchor boxes.

[0074] The Conv structure plays a role in feature extraction, the DLA-ELAN structure improves the utilization rate of model parameters, the MPConv structure reduces the risk of feature loss, the CSPNet effectively alleviates the complex mapping problem of the serial deep learning model and reduces the training difficulty, and the RepConv structure realizes the speed-up function in the model reasoning process. Through the organic combination of various structures, an efficient target detection algorithm that takes into account lightweight, reasoning speed and accuracy is finally designed, thereby realizing efficient detection of engine appearance consistency.

[0075] In one embodiment of the present application, the first convolution result is processed by using a first DLA-ELAN structure to obtain a first processing result, including:

[0076] Performing convolution processing on the first convolution result to obtain a first intermediate convolution result, and performing segmentation processing on the first intermediate convolution result to obtain a first segmentation result and a second segmentation result;

[0077] The second segmentation result is processed by using the first DLA structure to obtain a first intermediate processing result;

[0078] Performing convolution processing on the first intermediate processing result to obtain a second intermediate convolution result, and processing the second intermediate convolution result using a second DLA structure to obtain a second intermediate processing result;

[0079] Performing convolution processing on the second intermediate processing result to obtain a third intermediate convolution result;

[0080] The first intermediate convolution result, the first segmentation result, the second intermediate convolution result and the second intermediate processing result are combined to obtain the first processing result.

[0081] Specifically, the DLA-ELAN structure is as follows Figure 3As shown, no further details will be given here.

[0082] Step S103, determining whether the engine is abnormal based on the position confidence and the type confidence of the target part, wherein the engine abnormality indicates that the target part does not exist on the engine, or the target part on the engine is located at an incorrect position.

[0083] In the above steps, the shooting device is controlled to take multiple engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and the engine image is obtained, so as to ensure that the subsequent operation trajectory of the shooting device can completely capture all the appearance surfaces of the engine, and perform data collection, and use the image recognition model to predict the engine image to obtain multiple resolution images of the target part and the corresponding position confidence and type confidence. According to the position confidence and type confidence of the target part, it is determined whether the engine is abnormal, thereby realizing an efficient target detection algorithm that takes into account lightweight, reasoning speed and accuracy, and realizing efficient detection of the consistency of the engine appearance, thereby solving the problem that the existing solution needs to add manual auxiliary detection when detecting whether the parts outside the engine are complete, resulting in poor detection efficiency.

[0084] Determining whether the engine is abnormal according to the position confidence and the type confidence of the target part in step S103 includes:

[0085] In the case where the type corresponding to the maximum value of all the type confidences of the target part is the target type, determining the type of the target part to be the target type;

[0086] In the case where the position confidence of the target part is greater than or equal to a confidence threshold, determining that the engine is normal;

[0087] When the position confidence of the target part is less than a confidence threshold, it is determined that the engine is abnormal.

[0088] Specifically, for example, if the confidence threshold is 90%, and the confidences of all the above types in the above target parts are respectively type A: 50%, type B: 60%, and type C: 70%, then the type of the above target part is determined to be type C.

[0089] When the position confidence of the above target part is 95%, it is determined that the engine is normal and the part is correctly installed at the corresponding position of the engine.

[0090] When the position confidence of the above-mentioned target part is 75%, it is determined that the engine is abnormal and the part is not installed at the corresponding position of the engine.

[0091] like Figure 4 As shown in the figure, the training process of the model is as follows: after image preprocessing, the output value and label value are given, the model forward propagates to extract features, and compares the extracted features with the true coordinate values, and uses the loss function to calculate the loss value L between the target value and the output value. The loss value or the accurate value meets the requirements. If the requirements are met, the training ends. If the requirements are not met, the errors of each layer of the neural network are calculated, the error gradient is calculated according to the gradient derivative, and the weights of each neural network are updated according to the learning rate and gradient. I will not go into details here.

[0092] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the engine detection method of the present application will be described in detail below in conjunction with specific embodiments.

[0093] This embodiment relates to a specific engine detection method, such as Figure 5 As shown, including:

[0094] Acquire the initial flywheel housing image of the engine; use Hough circle detection technology and morphological threshold to determine the position in the initial flywheel housing image and obtain the current axis position;

[0095] When the current axis position is the center of the initial flywheel housing image, control the shooting device to take a plurality of engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and obtain the engine image; when the current axis position is not the center of the initial flywheel housing image, adjust the position of the shooting device until the current axis position is the center of the initial flywheel housing image;

[0096] The image recognition model is used to predict the engine image, and multiple resolution images of the target parts and the corresponding position confidence and type confidence are obtained;

[0097] Specifically, the engine image is convolved three times in sequence to obtain a first convolution result; the first convolution result is processed using the first DLA-ELAN structure to obtain a first processing result; the first processing result is processed using the first MPConv structure and the second DLA-ELAN structure in sequence to obtain a second processing result; the second processing result is processed using the second MPConv structure, the third DLA-ELAN structure and the CSPNet structure in sequence to obtain a third processing result.

[0098] The first processing result is convolved to obtain a second convolution result, and the second processing result is convolved to obtain a third convolution result; the third processing result is sequentially convolved and up-sampled to obtain a fourth processing result, and the up-sampling represents an image magnification process; the fourth processing result and the third convolution result are merged to obtain a first merged result; the first merged result is processed using the fourth DLA-ELAN structure to obtain a fifth processing result; the fifth processing result is sequentially convolved and up-sampled to obtain a sixth processing result; the second convolution result and the sixth processing result are merged to obtain a second merged result.

[0099] sequentially using the fifth DLA-ELAN structure and the third MPConv structure to process the second merging result to obtain a seventh processing result; merging the fifth processing result and the seventh processing result to obtain a third merging result;

[0100] The third merged result is processed in sequence using the sixth DLA-ELAN structure and the fourth MPConv structure to obtain an eighth processing result; the second processing result is processed in sequence using the second MPConv structure and the third DLA-ELAN structure to obtain a ninth processing result, and the eighth processing result and the ninth processing result are merged to obtain a fourth merged result; the fourth merged result is processed in sequence using the seventh DLA-ELAN structure and the first RepConv structure to obtain a first resolution image and corresponding position confidence and type confidence, and the third merged result is processed in sequence using the sixth DLA-ELAN structure and the second RepConv structure to obtain a second resolution image and corresponding position confidence and type confidence, and the second merged result is processed in sequence using the fifth DLA-ELAN structure and the third RepConv structure to obtain a third resolution image and corresponding position confidence and type confidence.

[0101] The first DLA-ELAN structure is used to process the first convolution result to obtain a first processing result, including: performing convolution processing on the first convolution result to obtain a first intermediate convolution result, and performing segmentation processing on the first intermediate convolution result to obtain a first segmentation result and a second segmentation result; using the first DLA structure to process the second segmentation result to obtain a first intermediate processing result; performing convolution processing on the first intermediate processing result to obtain a second intermediate convolution result, and using the second DLA structure to process the second intermediate convolution result to obtain a second intermediate processing result; performing convolution processing on the second intermediate processing result to obtain a third intermediate convolution result; and merging the first intermediate convolution result, the first segmentation result, the second intermediate convolution result and the second intermediate processing result to obtain the first processing result.

[0102] When the type corresponding to the maximum value of all type confidences in the target part is the target type, the type of the target part is determined to be the target type; when the position confidence of the target part is greater than or equal to the confidence threshold, the engine is determined to be normal; when the position confidence of the target part is less than the confidence threshold, the engine is determined to be abnormal, and the engine abnormality indicates that the target part does not exist on the engine, or the position of the target part on the engine is wrong.

[0103] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0104] The embodiment of the present application also provides a detection device for an engine. It should be noted that the detection device for an engine of the embodiment of the present application can be used to execute the detection method for an engine provided by the embodiment of the present application. The device is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived.

[0105] The following is an introduction to the engine detection device provided in the embodiments of the present application.

[0106] Figure 6 is a structural block diagram of an engine detection device provided according to an embodiment of the present application. Figure 6 As shown, the device comprises:

[0107] The first processing unit 61 is used to control the shooting device to shoot a plurality of engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and obtain the engine image;

[0108] The second processing unit 62 is used to predict the above engine image using an image recognition model to obtain multiple resolution images of the target part and corresponding position confidences and type confidences;

[0109] The third processing unit 63 is used to determine whether the engine is abnormal based on the position confidence and the type confidence of the target part. The engine abnormality indicates that the target part does not exist on the engine, or the position of the target part on the engine is wrong.

[0110] In the above-mentioned device, by controlling the shooting device to take a plurality of engine images around the above-mentioned engine with the bearing of the flywheel housing of the engine as the axis, and obtaining the above-mentioned engine image, it is ensured that the subsequent operation trajectory of the shooting device can completely shoot all the appearance surfaces of the engine, and perform data collection, and use the image recognition model to predict the above-mentioned engine image, and obtain multiple resolution images of the target part and the corresponding position confidence and type confidence. According to the above-mentioned position confidence and the above-mentioned type confidence of the above-mentioned target part, it is determined whether the above-mentioned engine is abnormal, thereby realizing an efficient target detection algorithm that takes into account lightweight, reasoning speed and accuracy, and realizing efficient detection of the consistency of the engine appearance, thereby solving the problem that the existing solution needs to add manual auxiliary detection when detecting whether the parts outside the engine are complete, resulting in poor detection efficiency.

[0111] In one embodiment of the present application, the image recognition model includes a first DLA-ELAN structure, a first MPConv structure, a second DLA-ELAN structure, a second MPConv structure, a third DLA-ELAN structure and a CSPNet structure, and the second processing unit includes a first processing module, a second processing module, a third processing module and a fourth processing module. In the process of using the image recognition model to predict the above-mentioned engine image and obtain multiple resolution images of the target part and the corresponding position confidence and type confidence,

[0112] The first processing module is used to perform three convolution processes on the engine image in sequence to obtain a first convolution result;

[0113] The second processing module is used to process the first convolution result by using the first DLA-ELAN structure to obtain a first processing result;

[0114] The third processing module is used to sequentially process the first processing result using the first MPConv structure and the second DLA-ELAN structure to obtain a second processing result;

[0115] The fourth processing module is used to sequentially process the second processing result using the second MPConv structure, the third DLA-ELAN structure and the CSPNet structure to obtain a third processing result.

[0116] In one embodiment of the present application, the image recognition model includes a fourth DLA-ELAN structure, the second processing unit includes a fifth processing module, a sixth processing module, a seventh processing module, an eighth processing module, a ninth processing module and a tenth processing module, and in the process of using the image recognition model to predict the engine image to obtain multiple resolution images of the target part and the corresponding position confidence and type confidence, the method further includes:

[0117] The fifth processing module is used for, after obtaining the third processing result, performing convolution processing on the first processing result to obtain a second convolution result, and performing convolution processing on the second processing result to obtain a third convolution result;

[0118] The sixth processing module is used to perform convolution processing and upsampling processing on the third processing result in sequence to obtain a fourth processing result, wherein the upsampling processing represents an image enlargement processing;

[0119] The seventh processing module is used to merge the fourth processing result and the third convolution result to obtain a first merged result;

[0120] The eighth processing module is used to process the first merging result by using the fourth DLA-ELAN structure to obtain a fifth processing result;

[0121] The ninth processing module is used to perform convolution processing and upsampling processing on the fifth processing result in sequence to obtain a sixth processing result;

[0122] The tenth processing module is used to merge the second convolution result and the sixth processing result to obtain a second merged result.

[0123] In one embodiment of the present application, the image recognition model includes a fifth DLA-ELAN structure, a third MPConv structure, a sixth DLA-ELAN structure, a fourth MPConv structure, a seventh DLA-ELAN structure, a first RepConv structure, a second RepConv structure, and a third RepConv structure, and the second processing unit includes an eleventh processing module, a twelfth processing module, a thirteenth processing module, a fourteenth processing module, and a fifteenth processing module. The image recognition model is used to predict the engine image to obtain multiple resolution images of the target part and corresponding position confidences and type confidences, including:

[0124] The eleventh processing module is used for, after obtaining the second merging result, sequentially using the fifth DLA-ELAN structure and the third MPConv structure to process the second merging result to obtain a seventh processing result;

[0125] The twelfth processing module is used to merge the fifth processing result and the seventh processing result to obtain a third merged result;

[0126] The thirteenth processing module is used to sequentially use the sixth DLA-ELAN structure and the fourth MPConv structure to process the third merging result to obtain an eighth processing result;

[0127] The fourteenth processing module is used to sequentially use the second MPConv structure and the third DLA-ELAN structure to process the second processing result to obtain a ninth processing result, and merge the eighth processing result and the ninth processing result to obtain a fourth merged result;

[0128] The fifteenth processing module is used to sequentially use the seventh DLA-ELAN structure and the first RepConv structure to process the fourth merged result to obtain a first resolution image and corresponding position confidence and type confidence, and at the same time sequentially use the sixth DLA-ELAN structure and the second RepConv structure to process the third merged result to obtain a second resolution image and corresponding position confidence and type confidence, and at the same time sequentially use the fifth DLA-ELAN structure and the third RepConv structure to process the second merged result to obtain a third resolution image and corresponding position confidence and type confidence.

[0129] In one embodiment of the present application, the second processing module includes a first processing submodule, a second processing submodule, a third processing submodule, a fourth processing submodule and a fifth processing submodule.

[0130] The first processing submodule is used to perform convolution processing on the first convolution result to obtain a first intermediate convolution result, and to perform segmentation processing on the first intermediate convolution result to obtain a first segmentation result and a second segmentation result;

[0131] The second processing submodule is used to process the second segmentation result by using the first DLA structure to obtain a first intermediate processing result;

[0132] The third processing submodule is used to perform convolution processing on the first intermediate processing result to obtain a second intermediate convolution result, and use a second DLA structure to process the second intermediate convolution result to obtain a second intermediate processing result;

[0133] The fourth processing submodule is used to perform convolution processing on the second intermediate processing result to obtain a third intermediate convolution result;

[0134] The fifth processing submodule is used to merge the first intermediate convolution result, the first segmentation result, the second intermediate convolution result and the second intermediate processing result to obtain the first processing result.

[0135] In one embodiment of the present application, the third processing unit includes a first determining module, a second determining module and a third determining module.

[0136] The first determination module is used for determining that the type of the target part is the target type when the type corresponding to the maximum value among all the type confidences of the target part is the target type;

[0137] The second determination module is used to determine that the engine is normal when the position confidence of the target part is greater than or equal to the confidence threshold;

[0138] The third determination module is used to determine that the engine is abnormal when the position confidence of the target part is less than a confidence threshold.

[0139] In one embodiment of the present application, the first processing unit includes an acquisition module, a sixteenth processing module, a seventeenth processing module and an eighteenth processing module.

[0140] The acquisition module is used to acquire the initial flywheel housing image of the engine;

[0141] The sixteenth processing module is used to determine the position of the initial flywheel housing image by using Hough circle detection technology and morphological threshold to obtain the current axis position;

[0142] The seventeenth processing module is used for controlling the photographing device to take a plurality of images of the engine around the engine with the bearing of the flywheel housing of the engine as the axis when the current axis position is the center of the initial flywheel housing image;

[0143] The eighteenth processing module is used to adjust the position of the above-mentioned shooting device when the above-mentioned current axis center position is not the center of the above-mentioned initial flywheel housing image, until the above-mentioned current axis center position is the center of the above-mentioned initial flywheel housing image.

[0144] The engine detection device includes a processor and a memory. The first processing unit, the second processing unit and the third processing unit are stored in the memory as program units, and the processor executes the program units stored in the memory to implement corresponding functions. The modules are all located in the same processor; or, the modules are located in different processors in any combination.

[0145] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the kernel parameters can be adjusted to solve the problem that the existing solution needs to add manual auxiliary detection when detecting whether the external parts of the engine are complete, resulting in poor detection efficiency.

[0146] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0147] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the engine detection method.

[0148] An embodiment of the present invention provides a processor, and the processor is used to run a program, wherein the engine detection method is executed when the program is run.

[0149] An embodiment of the present invention provides a device, the device includes a processor, a memory, and a program stored in the memory and executable on the processor, and the processor implements at least the following steps when executing the program: controlling the shooting device to take a plurality of engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and acquiring the engine image; using an image recognition model to predict the engine image, and obtain a plurality of resolution images of the target part and the corresponding position confidence and type confidence; determining whether the engine is abnormal according to the position confidence and type confidence of the target part, and the abnormal engine indicates that the target part does not exist on the engine, or the target part on the engine is in the wrong position. The device in this article can be a server, PC, PAD, mobile phone, etc.

[0150] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that initializes at least the following method steps: controlling a shooting device to take multiple engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and acquiring the engine image; using an image recognition model to predict the engine image to obtain multiple resolution images of the target part and corresponding position confidence and type confidence; determining whether the engine is abnormal based on the position confidence and type confidence of the target part, wherein the engine abnormality indicates that the target part does not exist on the engine, or the target part on the engine is in an incorrect position.

[0151] The present application also provides an engine detection system, which includes: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include methods for executing any of the above methods. By controlling the shooting device to take the bearing of the flywheel housing of the engine as the axis, taking multiple engine images around the engine, and acquiring the engine image, it is ensured that the subsequent running trajectory of the shooting device can completely capture all the appearance surfaces of the engine, and data collection is performed, and the engine image is predicted using an image recognition model to obtain multiple resolution images of the target part and the corresponding position confidence and type confidence. According to the position confidence and type confidence of the target part, it is determined whether the engine is abnormal, thereby realizing an efficient target detection algorithm that takes into account lightweight, reasoning speed and accuracy, and realizing efficient detection of the consistency of the engine appearance, thereby solving the problem that the existing solution needs to add manual auxiliary detection when detecting whether the parts outside the engine are complete, resulting in poor detection efficiency.

[0152] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0153] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0154] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0155] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0156] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0157] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0158] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0159] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0160] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0161] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0162] 1) The engine detection method of the present application controls the shooting device to take multiple engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and obtains the engine image, thereby ensuring that the subsequent operation trajectory of the shooting device can completely capture all the appearance surfaces of the engine, and collect data, and use the image recognition model to predict the engine image to obtain multiple resolution images of the target part and the corresponding position confidence and type confidence. According to the position confidence and type confidence of the target part, it is determined whether the engine is abnormal, thereby realizing an efficient target detection algorithm that takes into account lightweight, reasoning speed and accuracy, and realizing efficient detection of the consistency of the engine appearance, thereby solving the problem that the existing solution needs to add manual auxiliary detection when detecting whether the external parts of the engine are complete, resulting in poor detection efficiency.

[0163] 2) The engine detection device of the present application controls the shooting device to take multiple engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and obtains the above-mentioned engine image, thereby ensuring that the subsequent operation trajectory of the shooting device can completely capture all the appearance surfaces of the engine, and perform data collection, and use the image recognition model to predict the above-mentioned engine image to obtain multiple resolution images of the target part and the corresponding position confidence and type confidence. According to the above-mentioned position confidence and the above-mentioned type confidence of the above-mentioned target part, it is determined whether the above-mentioned engine is abnormal, thereby realizing an efficient target detection algorithm that takes into account lightweight, reasoning speed and accuracy, and realizing efficient detection of the consistency of the engine appearance, thereby solving the problem that the existing solution needs to add manual auxiliary detection when detecting whether the external parts of the engine are complete, resulting in poor detection efficiency.

[0164] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. 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 detecting an engine, characterized in that: include: Controlling the photographing device to take a plurality of engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and acquiring the engine images; Using an image recognition model to predict the engine image, obtaining multiple resolution images of the target part and corresponding position confidences and type confidences; Determining whether the engine is abnormal according to the position confidence and the type confidence of the target part, wherein the engine abnormality indicates that the target part does not exist on the engine, or the target part on the engine is located at an incorrect position; The image recognition model includes a first DLA-ELAN structure, a first MPConv structure, a second DLA-ELAN structure, a second MPConv structure, a third DLA-ELAN structure and a CSPNet structure. In the process of using the image recognition model to predict the engine image and obtain multiple resolution images of the target part and the corresponding position confidence and type confidence, the process includes: Performing convolution processing on the engine image three times in sequence to obtain a first convolution result; Processing the first convolution result by using the first DLA-ELAN structure to obtain a first processing result; sequentially using the first MPConv structure and the second DLA-ELAN structure to process the first processing result to obtain a second processing result; sequentially using the second MPConv structure, the third DLA-ELAN structure, and the CSPNet structure to process the second processing result to obtain a third processing result; The first convolution result is processed using a first DLA-ELAN structure to obtain a first processing result, including: Performing convolution processing on the first convolution result to obtain a first intermediate convolution result, and performing segmentation processing on the first intermediate convolution result to obtain a first segmentation result and a second segmentation result; Processing the second segmentation result using the first DLA structure to obtain a first intermediate processing result; Performing convolution processing on the first intermediate processing result to obtain a second intermediate convolution result, and processing the second intermediate convolution result using a second DLA structure to obtain a second intermediate processing result; Performing convolution processing on the second intermediate processing result to obtain a third intermediate convolution result; Merging the first intermediate convolution result, the first segmentation result, the second intermediate convolution result, the second intermediate processing result, and the third intermediate convolution result to obtain the first processing result; The DLA-ELAN structure is used to improve the utilization of model parameters, the MPConv structure is used to reduce the risk of feature loss, the CSPNet is used to alleviate the complex mapping problem of the serial deep learning model and reduce the difficulty of training, and the RepConv structure is used to achieve the speed-up function in the model reasoning process.

2. The method according to claim 1, characterized in that The image recognition model includes a fourth DLA-ELAN structure, and in the process of using the image recognition model to predict the engine image to obtain multiple resolution images of the target part and the corresponding position confidence and type confidence, it includes: After obtaining the third processing result, performing convolution processing on the first processing result to obtain a second convolution result, and performing convolution processing on the second processing result to obtain a third convolution result; Performing convolution processing and upsampling processing on the third processing result in sequence to obtain a fourth processing result, wherein the upsampling processing represents enlarging the image; Combining the fourth processing result and the third convolution result to obtain a first combined result; Processing the first merging result by using the fourth DLA-ELAN structure to obtain a fifth processing result; performing convolution processing and upsampling processing on the fifth processing result in sequence to obtain a sixth processing result; The second convolution result and the sixth processing result are combined to obtain a second combined result.

3. The method according to claim 2, characterized in that The image recognition model includes a fifth DLA-ELAN structure, a third MPConv structure, a sixth DLA-ELAN structure, a fourth MPConv structure, a seventh DLA-ELAN structure, a first RepConv structure, a second RepConv structure, and a third RepConv structure. The image recognition model is used to predict the engine image to obtain multiple resolution images of the target part and corresponding position confidences and type confidences, including: After obtaining the second merging result, sequentially using the fifth DLA-ELAN structure and the third MPConv structure to process the second merging result to obtain a seventh processing result; Combining the fifth processing result and the seventh processing result to obtain a third combined result; sequentially using the sixth DLA-ELAN structure and the fourth MPConv structure to process the third merging result to obtain an eighth processing result; sequentially using the second MPConv structure and the third DLA-ELAN structure to process the second processing result to obtain a ninth processing result, and merging the eighth processing result and the ninth processing result to obtain a fourth merging result; The seventh DLA-ELAN structure and the first RepConv structure are used in sequence to process the fourth merged result to obtain a first resolution image and a corresponding position confidence and a type confidence. At the same time, the sixth DLA-ELAN structure and the second RepConv structure are used in sequence to process the third merged result to obtain a second resolution image and a corresponding position confidence and a type confidence. At the same time, the fifth DLA-ELAN structure and the third RepConv structure are used in sequence to process the second merged result to obtain a third resolution image and a corresponding position confidence and a type confidence.

4. The method according to claim 1, characterized in that: Determining whether the engine is abnormal according to the position confidence and the type confidence of the target part includes: In a case where the type corresponding to the maximum value among all the type confidences in the target part is the target type, determining the type of the target part to be the target type; When the position confidence of the target part is greater than or equal to a confidence threshold, determining that the engine is normal; In a case where the position confidence of the target part is less than a confidence threshold, it is determined that the engine is abnormal.

5. The method according to claim 1, characterized in that Controlling the photographing device to take a plurality of images of the engine around the engine with the bearing of the flywheel housing of the engine as the axis, including: Acquiring an initial flywheel housing image of the engine; Using Hough circle detection technology and morphological threshold to determine the position of the initial flywheel housing image, and obtain the current axis position; When the current axis position is the center of the initial flywheel housing image, controlling the shooting device to shoot a plurality of images of the engine around the engine with the bearing of the flywheel housing of the engine as the axis; When the current axis position is not the center of the initial flywheel housing image, the position of the photographing device is adjusted until the current axis position is the center of the initial flywheel housing image.

6. An engine detection device, characterized in that: include: A first processing unit is used to control the shooting device to shoot a plurality of engine images around the engine with the bearing of the flywheel housing of the engine as the axis, and acquire the engine images; A second processing unit is used to predict the engine image using an image recognition model to obtain multiple resolution images of the target part and corresponding position confidences and type confidences; a third processing unit, configured to determine whether the engine is abnormal according to the position confidence and the type confidence of the target part, wherein the engine abnormality indicates that the target part does not exist on the engine, or the target part on the engine is located at an incorrect position; The image recognition model includes a first DLA-ELAN structure, a first MPConv structure, a second DLA-ELAN structure, a second MPConv structure, a third DLA-ELAN structure and a CSPNet structure, and the second processing unit includes a first processing module, a second processing module, a third processing module and a fourth processing module. In the process of using the image recognition model to predict the engine image and obtain multiple resolution images of the target part and the corresponding position confidence and type confidence, The first processing module is used to perform three convolution processes on the engine image in sequence to obtain a first convolution result; The second processing module is used to process the first convolution result by using the first DLA-ELAN structure to obtain a first processing result; The third processing module is used to sequentially process the first processing result using the first MPConv structure and the second DLA-ELAN structure to obtain a second processing result; The fourth processing module is used to sequentially process the second processing result using the second MPConv structure, the third DLA-ELAN structure and the CSPNet structure to obtain a third processing result; The second processing module includes a first processing submodule, a second processing submodule, a third processing submodule, a fourth processing submodule and a fifth processing submodule. The first processing submodule is used to perform convolution processing on the first convolution result to obtain a first intermediate convolution result, and perform segmentation processing on the first intermediate convolution result to obtain a first segmentation result and a second segmentation result; The second processing submodule is used to process the second segmentation result by using the first DLA structure to obtain a first intermediate processing result; The third processing submodule is used to perform convolution processing on the first intermediate processing result to obtain a second intermediate convolution result, and use a second DLA structure to process the second intermediate convolution result to obtain a second intermediate processing result; The fourth processing submodule is used to perform convolution processing on the second intermediate processing result to obtain a third intermediate convolution result; The fifth processing submodule is used to merge the first intermediate convolution result, the first segmentation result, the second intermediate convolution result, the second intermediate processing result and the third intermediate convolution result to obtain the first processing result; The DLA-ELAN structure is used to improve the utilization of model parameters, the MPConv structure is used to reduce the risk of feature loss, the CSPNet is used to alleviate the complex mapping problem of the serial deep learning model and reduce the difficulty of training, and the RepConv structure is used to achieve the speed-up function in the model reasoning process.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 5.

8. An engine detection system, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs are used to execute the method described in any one of claims 1 to 5.

Images