A chain split pin defect intelligent detection method and system

Abstract

The present invention discloses a method and system for intelligently detecting defects of chain cotter pins, the method comprising the following steps: obtaining an image of a cotter pin type chain joint to be detected; judging whether the cotter pin is contained in the image of the chain joint; if not, judging that there is a defect of missing chain cotter pin in the image to be detected; if contained, performing cotter pin key point detection based on the image of the region where the cotter pin is located to obtain multiple key point positions; according to the multiple key point positions, calculating the opening angle of the cotter pin tail to judge whether the cotter pin is normal. The present invention can quickly and accurately judge the state of the cotter pin.

Description

Technical Field

[0001] The present invention belongs to the technical field of transmission component defect detection, and in particular relates to a chain cotter pin defect intelligent detection method and system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Chain is a common mechanical transmission element, usually composed of a series of chain links connected by pins. Chains can be used in a variety of application scenarios, including power transmission, lifting and handling. For example, to solve the problem of parking difficulties, building a three-dimensional parking garage based on chain transmission has great advantages in effectively improving the utilization rate of parking spaces. Chains are usually composed of inner chain plates, outer chain plates, pins, rollers, sleeves and other parts. In order to achieve the connection of multiple chain sections, cotter pin type chain joints are a commonly used connection form. In the chain of the three-dimensional parking garage, the cotter pin is mainly used to prevent the chain from loosening or falling off between the car plate and the chain. However, the chain is affected by the vibration and impact generated during operation for a long time, which makes the cotter pin have the risk of loosening, breaking, missing, etc., which poses a hidden danger to the safe operation of mechanical equipment such as the three-dimensional parking garage, and seriously threatens the safety of life and property of users.

[0004] Currently, manual inspection is mainly used to detect defects such as missing cotter pins and insufficient opening angles of chains. However, traditional manual inspection has the disadvantages of being time-consuming, labor-intensive, highly subjective, and inefficient. In particular, it is difficult for manual inspection to accurately identify defects such as insufficient opening angles of cotter pins. Summary of the invention

[0005] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a chain cotter pin defect intelligent detection method and system, which can quickly and accurately determine the cotter pin status.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] A chain split pin defect intelligent detection method comprises the following steps:

[0008] Acquire the image of the cotter pin type chain joint to be inspected;

[0009] Determine whether the chain joint image contains a split pin:

[0010] If not, it is determined that there is a defect of missing chain cotter pin in the image to be detected;

[0011] If included, based on the image of the area where the cotter pin is located, perform cotter pin key point detection to obtain multiple key point positions; based on the multiple key point positions, calculate the opening angle of the cotter pin tail to determine whether the cotter pin is normal.

[0012] In some embodiments, whether the chain joint image contains a split pin is determined based on a split pin detection network; the split pin detection network includes multiple feature extraction layers and bidirectional feature pyramids connected in sequence, the number of feature extraction layers is the same as the number of upsampling layers or downsampling layers in the bidirectional feature pyramid, denoted as n; wherein the output end of the i-th feature extraction layer is connected to the input end of the n-i+1-th upsampling layer.

[0013] In some embodiments, cotter pin key point detection is performed based on a key point detection network; the key point detection network includes multiple dual convolutional layers and fully connected layers connected in sequence, and the output end of each dual convolutional layer is connected to the fully connected layer; wherein a multi-scale context feature module is embedded between the two convolutional layers of the first dual convolutional layer, and the multi-scale context feature module includes multiple feature extraction branches, and the multiple feature extraction branches are connected to the next convolutional layer via channels, and the convolution block expansion rate on each branch is different, and the corresponding channel weight size is different.

[0014] In some embodiments, the multi-scale contextual feature module includes three feature extraction branches, which are respectively used to detect the key points of the cotter pin as a whole, the head area and the two tail areas. The weight of the branch corresponding to the channel used to detect the two tail areas is the largest, and the weight of the branch corresponding to the channel used to detect the head area is the smallest.

[0015] In some embodiments, calculating the opening angle of the cotter pin tail specifically includes: performing curve fitting of the two tails based on key points corresponding to the two tail areas of the cotter pin passing through the pin hole, and calculating the opening angle of the cotter pin tail based on the two fitting curves.

[0016] In some embodiments, the curve fitting adopts Bezier curve fitting.

[0017] One or more embodiments provide a chain cotter pin defect intelligent detection system, including:

[0018] An image acquisition module is configured to acquire an image of a cotter pin type chain joint to be inspected;

[0019] The split pin detection module is configured to determine whether the chain joint image contains a split pin:

[0020] The defect detection module is configured to, if the cotter pin is not included, determine that there is a defect of missing chain cotter pin in the image to be detected; if the cotter pin is included, perform cotter pin key point detection based on the image of the area where the cotter pin is located to obtain multiple key point positions; based on the multiple key point positions, calculate the opening angle of the cotter pin tail to determine whether the cotter pin is normal.

[0021] One or more embodiments provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the chain cotter pin defect intelligent detection method when executing the program.

[0022] One or more embodiments provide a computer-readable storage medium having a computer program stored thereon, which implements the chain cotter pin defect intelligent detection method when executed by a processor.

[0023] One or more embodiments provide a computer program product, including computer executable instructions, which, when executed by a processor, implement the chain cotter pin defect intelligent detection method.

[0024] One or more of the above technical solutions have the following beneficial effects:

[0025] By dividing the cotter pin defect detection into three stages: cotter pin positioning, key point detection and defect detection, the two major defects of missing cotter pin and insufficient cotter pin opening angle can be accurately identified, thereby realizing intelligent detection of chain cotter pins and reducing safety hazards caused by chain cotter pin defects during the transmission process of large mechanical equipment.

[0026] In the split pin detection stage, a lateral connection is established between the feature extraction layer and the bidirectional feature pyramid to integrate multi-scale semantic information and position information, thereby improving the detection accuracy of the split pin as a small target.

[0027] By setting multiple key points covering all shape features of the split pin, convolution kernels and channel weights are designed specifically for the overall and different areas of the split pin in the key point detection stage, thereby improving the detection accuracy of the split pin key points. In addition, in addition to being able to obtain the coordinates of these key points, recognition based on key point detection can also verify the split pin detection results.

[0028] Through intelligent detection, the chain cotter pin images can be processed in batches, which improves the detection efficiency. Compared with manual detection, the inspection results are more accurate and objective, and there is no need for inspectors to go to the site to check the status of the cotter pins. To a certain extent, it reduces the consumption of manpower and material resources, and also reduces the risk of safety accidents during the inspection process. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0030] Figure 1 A schematic diagram of a process flow of an intelligent detection method for chain cotter pin defects in one or more embodiments of the present invention;

[0031] Figure 2 A schematic diagram of a split pin type chain joint in one or more embodiments of the present invention;

[0032] Figure 3 A schematic diagram of a split pin detection network in one or more embodiments of the present invention;

[0033] Figure 4 A schematic diagram focusing on the split pin area in one or more embodiments of the present invention;

[0034] Figure 5 A schematic diagram of key points of the image of the split pin region in one or more embodiments of the present invention;

[0035] Figure 6 A schematic diagram of a key point detection network in one or more embodiments of the present invention;

[0036] Figure 7 A schematic diagram of key point detection results in one or more embodiments of the present invention;

[0037] Figure 8 It is a schematic diagram of calculating the opening angle of the tail of the cotter pin in one or more embodiments of the present invention. DETAILED DESCRIPTION

[0038] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0039] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0040] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0041] The present invention combines the deep learning algorithm with the cotter pin defect detection of the stereo garage chain, which can accurately determine the state of the cotter pin, save a lot of manpower and material resources compared with manual detection, and the detection result is more accurate and objective. Since the cotter pin area of ​​the stereo garage chain occupies a small proportion of the pixels of the entire image (the pixel size of the cotter pin area is usually 30×30, and the resolution of the image acquisition camera can reach 3600×2800), it is impossible to directly detect the relevant information. Therefore, the present invention divides the cotter pin defect intelligent detection method into three parts. First, the image containing the cotter pin of the stereo garage chain is sent to the target detection network to focus on the key area of ​​the cotter pin; the located cotter pin area information image is sent to the key point detection network to obtain the key point coordinates; the cotter pin angle is obtained according to the key point coordinates. The angle calculation network is input to calculate the angle of the cotter pin. In this process, the defects of missing cotter pins and insufficient cotter pin opening angles are automatically identified, thereby realizing the intelligent detection of cotter pin defects in the stereo garage chain.

[0042] One or more embodiments of the present invention provide a chain split pin defect intelligent detection method, such as Figure 1 As shown, the following steps are included:

[0043] Step 1: Obtain the image of the cotter pin type chain joint to be inspected.

[0044] Specifically, an industrial camera is used to collect images. Figure 2 An example of an acquired joint image is shown.

[0045] For a stereo garage, as a specific implementation method, a camera can be set on the steel beam of the stereo garage. During the chain retraction and extension process, one or more chains within the field of view are photographed, and the split pin type chain joints are identified. The split pin type chain joints can be identified using existing identification technology, which is not determined here.

[0046] As another specific implementation method, a track can be built on the structural steel beam of the stereo garage in cooperation with a stereo garage inspection robot. The robot walks on the track of the steel beam and sets an inspection path according to the location of the joint on the chain. The inspection path includes multiple detection points corresponding to the joint position. When the chain is reset, the inspection robot obtains the joint image according to the inspection path. The track robot should keep the speed constant when moving, and keep the focal length of the camera as constant as possible, so that the acquired image parameters are consistent, which is convenient for subsequent processing.

[0047] In actual engineering scenes, as a small-sized target, the split pin occupies a very small proportion in the collected image, and the positioning of the split pin is relatively difficult. In addition, non-target interferences such as rivets, cables, and engineering facilities are common in inspection scenes, which seriously affect the extraction accuracy of the split pin target.

[0048] This embodiment first guides the neural network to locate the area where the cotter pin is located through target detection, locates the position information of the cotter pin area in the image, and filters out background interference such as cables and rivets in the inspection scene.

[0049] Step 2: Determine whether the chain joint image contains a cotter pin. If not, determine whether there is a defect of missing chain cotter pin in the image to be detected. If yes, execute step 3.

[0050] In some embodiments, the presence of a split pin in the chain joint image is determined based on a pre-trained split pin detection network. The acquired image of the chain joint to be detected in the stereoscopic parking garage is input into the trained target detection network to determine whether the image to be detected contains information about the split pin area, and a cropping frame locating the split pin area is output, which will be fed to the subsequent key point detection network.

[0051] As a specific implementation method, Figure 3 As shown, the split pin detection network includes a feature extraction module, a feature fusion module and a target prediction module connected in sequence. The feature extraction module includes a plurality of feature extraction layers connected in sequence, each of which includes one or more convolutional layers and a residual block. The feature fusion module adopts a bidirectional feature pyramid (Feature Pyramid Networks, FPN), which includes a plurality of upsampling layers from top to bottom and a plurality of downsampling layers from bottom to top. The target prediction module adopts non-maximum suppression (NMS) and least squares strategy to fit an optimal cropping frame from a large number of detection results. The cropping frame includes the split pin information so that it can be subsequently input into the key point detection network for key point prediction.

[0052] The number of upsampling layers and downsampling layers of the feature extraction layer and the bidirectional feature pyramid is the same, denoted as n. Specifically, the output end of the first feature extraction layer is connected to the input end of the nth upsampling layer, and the output end of the nth upsampling layer is connected to the input end of the first downsampling layer; the output end of the second feature extraction layer is connected to the input end of the n-1th upsampling layer, and the output end of the n-1th upsampling layer is connected to the input end of the second downsampling layer; ..., and so on, the output end of the nth feature extraction layer is connected to the first upsampling layer, and the output end of the first upsampling layer is connected to the input end of the nth downsampling layer, such as Figure 3 By establishing lateral connections between the feature extraction layer and the upsampling layer, and between the upsampling layer and the downsampling layer, feature fusion is strengthened, and multi-scale semantic information and position information are integrated to improve the detection accuracy of small objects without sacrificing expression ability, speed, and computing resources.

[0053] To train the split pin detection network, for the training image containing the split pin, use Labelme software to select the split pin area in the captured image to obtain a json file corresponding to the input image, convert the json file into a corresponding txt file through the software, and input the captured image and the corresponding txt file into the network to train the split pin detection network.

[0054] like Figure 4 As shown, the image to be detected is sent to the split pin detection network to identify the area where the split pin is located. The step 2 clarifies the conditions for inputting the key point detection network. Only when the split pin target detection network outputs the split pin area information image will it enter the subsequent operation. Otherwise, it will directly determine that the split pin in the original input image is missing, thereby improving the detection efficiency.

[0055] Step 3: Based on the image of the area where the cotter pin is located, perform cotter pin key point detection to obtain multiple key point positions.

[0056] In some embodiments, a pre-trained key point detection network is used to perform key point detection on the cotter pin region image. In this embodiment, a total of 13 key points are set, specifically, Figure 5 As shown, the cotter pin is divided into three regions including the head and two tails on both sides of the pin hole. The head region is recorded as the first region, and the two tail regions are recorded as the second region and the third region. Among them, the first region has 5 key points, recorded as key points 0-4, the second region has 4 key points, recorded as key points 5-8, and the third region has 4 key points, recorded as key points 9-12.

[0057] As a specific implementation method, Figure 6 As shown, the key point detection network includes multiple double convolutional layers and fully connected layers connected in sequence, and the output end of each double convolutional layer is connected to the fully connected layer; wherein, each double convolutional layer, as the name implies, includes two convolutional layers, and a multi-scale context feature module is embedded between the two convolutional layers of the first double convolutional layer, and the multi-scale context feature module includes three feature extraction branches, and the three feature extraction branches are connected to the next convolutional layer via channels, and the convolution blocks on each branch have different expansion rates and corresponding channel weights. The fully connected layer is used to weightedly fuse the features output by each double convolutional layer to predict the key point coordinates.

[0058] Specifically, an 8-layer convolutional neural network is used, and its downsampling process includes four levels. Each level contains a double convolution process, which consists of two 3×3 convolution kernels and a nonlinear activation unit, and the number of convolution kernel channels in each level increases from top to bottom. The upper and lower levels are processed by a 2×2 maximum pooling layer, so that the feature map size of the next layer input is half of the output of the previous layer. Among them, in order to improve the feature extraction ability of the neural network for the key points of the open pin, a multi-scale context feature module is embedded between the two convolution layers of the first level to balance the information ratio of the key points of the open pin in the feature maps of different sizes. The three branches of the multi-scale context feature module are used to detect the key points of the open pin as a whole, the head area and the two tail areas, respectively. The weight of the corresponding channel of the branch used to detect the two tail areas is the largest, and the weight of the corresponding channel of the branch used to detect the head area is the smallest. Preferably, the multi-scale context feature module sets convolution blocks with expansion rates of 3, 9, and 18 respectively, and uses convolution blocks with different expansion rates to fuse the corresponding feature map information of the open pin key points into the channel information, guiding the network to focus on the area where the key points may exist on the open pin. The branch with an expansion rate of 3 corresponds to a convolution kernel size of 3*3, which is mainly used to extract the open pin pixels in the cropping frame. At the same time, since the receptive field of this layer is large, it has certain advantages in predicting the open pin as a whole. The branch with an expansion rate of 9 corresponds to a convolution kernel size of 5*3, which is mainly used to detect the 5 key points distributed on the head of the open pin (the first area). The branch with an expansion rate of 18 corresponds to a convolution kernel size of 3*5, which is mainly used for the detection of key points of the foot (the second and third areas). The feature maps obtained by the three branches are compressed into a one-dimensional matrix, and then the importance of each channel is assigned. The branch with an expansion rate of 3 corresponds to a channel weight of 0.3, the branch with an expansion rate of 9 corresponds to a channel weight of 0.2, and the branch with an expansion rate of 18 corresponds to a channel weight of 0.5. The three branches are linearly superimposed to obtain a comprehensive feature information map of the key points of the cotter pin, and then convolution operations are performed on subsequent levels.

[0059] The cotter pin area information image is input into the key point detection network, and the coordinates of 13 key points are output, such as Figure 7 shown.

[0060] The 13 key points set above cover all the shape features of the cotter pin. By performing key point detection, in addition to obtaining the coordinates of the 13 key points, the cotter pin detection results of step 2 can also be verified. If the detection fails, for example, the set key points cannot be detected, or the detection is incomplete, it is considered that the detection result of step 2 is incorrect and the cotter pin needs to be re-detected.

[0061] In order to train the key point detection network, Labelme software is used to mark the 13 key points of the split pin area image, and a corresponding json file is generated. The split pin area image and the corresponding json file are used to train the key point detection network.

[0062] At the same time, during the training phase, the network is evaluated and optimized by calculating the accuracy of the detection results. Specifically, the coordinates predicted by the network are compared with the actual marked coordinates, and the similarity between the two is calculated. This embodiment uses the OKS indicator, which has a value between (0,1). The larger the value, the higher the similarity, that is, the more accurate the prediction. The calculation formula is:

[0063]

[0064] Among them, i represents the i-th key point; d i Represents the Euclidean distance between the predicted value and the true value; v i Indicates whether the key point is visible; δ indicates that the key point is 1 when it is visible and 0 when it is invisible; k i represents a proportional constant; s represents the size of the object image. Generally, the similarity threshold can be set to 80%, and key points with accuracy less than 80% in the second and third regions are removed.

[0065] Step 4: Based on the positions of the multiple key points, calculate the opening angle of the cotter pin tail to determine whether the cotter pin is normal.

[0066] The insufficient opening angle of the cotter pin tail is the main reason for the cotter pin of the three-dimensional parking garage chain to fall off. This embodiment mainly detects this defect. The calculation of the opening angle of the cotter pin tail adopts the key points in the second area and the third area. The calculation of the opening angle of the cotter pin tail includes the following steps:

[0067] Step 4.1: Perform curve fitting based on the key points of the second region and the third region to obtain the fitting curves of the two tails of the split pin. The above curve fitting adopts Bezier curve fitting, and the fitted curves.

[0068] Step 4.2: If Figure 8 As shown, the tangent line A1 of the curve equation of the second area and the tangent line A2 of the curve equation of the third area are calculated respectively, and A1 and A2 intersect to form a cotter pin angle S1, and the angle of S1 is calculated by trigonometric functions.

[0069] Set the split pin angle threshold. Generally, the angle threshold is set to 40°. When the split pin angle is less than the set threshold, it is determined that the split pin of the chain in the image to be tested has a defect of insufficient split angle. When the split pin angle is greater than the set threshold, it is determined that the split pin of the chain in the image to be tested is in normal condition.

[0070] Based on the above method, the state of the split pin can be determined quickly, comprehensively, accurately and objectively, which can prevent the chain from loosening or breaking due to split pin defects, and at the same time save a lot of manpower and material resources and improve detection efficiency.

[0071] One or more embodiments of the present invention further provide a chain cotter pin defect intelligent detection system, comprising:

[0072] An image acquisition module is configured to acquire an image of a cotter pin type chain joint to be inspected;

[0073] The split pin detection module is configured to determine whether the chain joint image contains a split pin:

[0074] The defect detection module is configured to, if the cotter pin is not included, determine that there is a defect of missing chain cotter pin in the image to be detected; if the cotter pin is included, perform cotter pin key point detection based on the image of the area where the cotter pin is located to obtain multiple key point positions; based on the multiple key point positions, calculate the opening angle of the cotter pin tail to determine whether the cotter pin is normal.

[0075] One or more embodiments of the present invention also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the intelligent method for detecting chain cotter pin defects when executing the program.

[0076] In some embodiments, the chain split pin defect intelligent detection method can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed into an electronic device via a RAM and / or a ROM and / or a communication unit. When the computer program is loaded into the RAM and / or the ROM and executed by the electronic device, one or more steps of the method described above can be performed.

[0077] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A chain cotter pin defect intelligent detection method, characterized in that: The following steps are involved: Acquire the image of the cotter pin type chain joint to be inspected; Determine whether the chain joint image contains a split pin: If not, it is determined that there is a defect of missing chain cotter pin in the image to be detected; If included, based on the image of the area where the cotter pin is located, perform cotter pin key point detection to obtain multiple key point positions; calculate the opening angle of the cotter pin tail according to the multiple key point positions to determine whether the cotter pin is normal; Based on the split pin detection network, it is determined whether the chain joint image contains a split pin; the split pin detection network includes a plurality of feature extraction layers and a bidirectional feature pyramid connected in sequence, the number of layers of the feature extraction layer is the same as the number of upsampling layers or downsampling layers in the bidirectional feature pyramid, which is denoted as n; wherein the output end of the i-th feature extraction layer is connected to the input end of the n-i+1-th upsampling layer; The key point detection of the cotter pin is performed based on the key point detection network; the key point detection network includes a plurality of double convolutional layers and a fully connected layer connected in sequence, and the output end of each double convolutional layer is connected to the fully connected layer; wherein a multi-scale context feature module is embedded between the two convolutional layers of the first double convolutional layer, and the multi-scale context feature module includes a plurality of feature extraction branches, and the plurality of feature extraction branches are connected to the next convolutional layer via channels, and the expansion rate of the convolution block on each branch is different, and the corresponding channel weight size is different.

2. The chain cotter pin defect intelligent detection method according to claim 1, characterized in that: The multi-scale context feature module includes three feature extraction branches, which are respectively used to detect the key points of the cotter pin as a whole, the head area and the two tail areas. The weight of the branch corresponding to the channel used to detect the two tail areas is the largest, and the weight of the branch corresponding to the channel used to detect the head area is the smallest.

3. The chain cotter pin defect intelligent detection method according to any one of claims 1 to 2, characterized in that: Calculating the opening angle of the cotter pin tail specifically includes: performing curve fitting of the two tails based on key points corresponding to the two tail areas of the cotter pin passing through the pin hole, and calculating the opening angle of the cotter pin tail based on the two fitting curves.

4. The chain cotter pin defect intelligent detection method according to claim 3, characterized in that: The curve fitting adopts Bezier curve fitting.

5. A chain cotter pin defect intelligent detection system, characterized in that: include: An image acquisition module is configured to acquire an image of a cotter pin type chain joint to be inspected; The split pin detection module is configured to determine whether the chain joint image contains a split pin: The defect detection module is configured to determine that there is a defect of a missing chain cotter pin in the image to be detected if the cotter pin is not included; if the cotter pin is included, perform cotter pin key point detection based on the image of the area where the cotter pin is located to obtain multiple key point positions; calculate the opening angle of the cotter pin tail according to the multiple key point positions to determine whether the cotter pin is normal; Based on the split pin detection network, it is determined whether the chain joint image contains a split pin; the split pin detection network includes a plurality of feature extraction layers and a bidirectional feature pyramid connected in sequence, the number of layers of the feature extraction layer is the same as the number of upsampling layers or downsampling layers in the bidirectional feature pyramid, which is denoted as n; wherein the output end of the i-th feature extraction layer is connected to the input end of the n-i+1-th upsampling layer; The key point detection of the cotter pin is performed based on the key point detection network; the key point detection network includes a plurality of double convolutional layers and a fully connected layer connected in sequence, and the output end of each double convolutional layer is connected to the fully connected layer; wherein a multi-scale context feature module is embedded between the two convolutional layers of the first double convolutional layer, and the multi-scale context feature module includes a plurality of feature extraction branches, and the plurality of feature extraction branches are connected to the next convolutional layer via channels, and the expansion rate of the convolution block on each branch is different, and the corresponding channel weight size is different.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements the chain cotter pin defect intelligent detection method as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it implements the chain cotter pin defect intelligent detection method as described in any one of claims 1-4.

8. A computer program product comprising computer executable instructions, characterized in that: When the computer executable instructions are executed by the processor, the chain cotter pin defect intelligent detection method as described in any one of claims 1 to 4 is implemented.

Images