Vehicle target identification and detection method and system

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a vehicle target identification and detection method and system. Comprising the following steps: training an experimental data set by using a Yolo v5 target detection algorithm, and introducing a BIFPN bidirectional feature pyramid network and a CBAM-nSAM attention mechanism based on CBAM improvement; a GSConv structure based on depth separable convolution is introduced into a Yolo v5 target detection algorithm, convolution of a Neck part in the Yolo v5 network is replaced, and an EIoU loss function is used to replace a CIoU loss function in the Yolo v5 network; and training the improved Yolo v5 target detection algorithm by using the training set and the verification set to obtain a trained model and a trained weight, testing the trained model and the trained weight by using the test set, and further evaluating a result. According to the invention, based on the improvement of the BIFPN bidirectional feature fusion pyramid network and the CBAM-nSAM attention mechanism, a more accurate and more efficient vehicle identification method can be realized by using the existing hardware basis, and rapid and accurate vehicle identification statistics can be realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a vehicle target recognition and detection method and system. Background Art

[0002] With the continuous advancement of technology and the accelerated development of urbanization, the surge in the number of private cars has brought unprecedented challenges to the urban transportation system, especially during peak hours, when traffic congestion is particularly prominent.

[0003] In the field of intelligent traffic management, deep learning-based algorithms, such as SSD, Faster R-CNN and other target detection technologies, have been widely used in vehicle identification and traffic monitoring systems. Although these technologies perform well in certain application scenarios, they often face problems of insufficient detection accuracy or insufficient processing speed when dealing with real-time vehicle detection in complex traffic environments, making it difficult to meet the dual requirements of modern traffic management for high efficiency and high accuracy. Summary of the invention

[0004] In view of the shortcomings of the prior art described above, the purpose of the present invention is to provide a vehicle target recognition detection method and system to solve the problem of insufficient detection accuracy or insufficient processing speed, which is difficult to meet the dual requirements of modern traffic management for high efficiency and high accuracy.

[0005] To achieve the above objectives and other related objectives, the present invention provides a vehicle target recognition and detection method, comprising:

[0006] S1. Collect vehicle identification data, perform preprocessing on the vehicle identification data, i.e., data enhancement, to expand the data set, use the training set, validation set, and test set in the expanded data set to create an experimental data set, annotate the images of the experimental data set, and assign a label to each annotation;

[0007] S2. Use the Yolo v5 target detection algorithm to train the experimental data set, improve the Yolo v5 target detection algorithm based on the single-stage detection algorithm Yolo v5 network, and introduce the BIFPN bidirectional feature pyramid network and the CBAM-nSAM attention mechanism based on the CBAM improvement;

[0008] S3, continue to optimize the improved Yolo v5 target detection algorithm, introduce the GSConv structure based on deep separable convolution into the Yolo v5 target detection algorithm, replace the convolution of the Neck part in the Yolo v5 network, and use the EIoU loss function to replace the CIoU loss function in the Yolo v5 network;

[0009] S4. Use the training set and validation set to train the improved Yolo v5 target detection algorithm to obtain the trained model and weights. Use the test set to test the trained model and weights to evaluate the results.

[0010] In one embodiment of the present invention, the ratio of the training set, the validation set, and the test set is 7:1.5:1.5.

[0011] In one embodiment of the present invention, the data enhancement in step S1 includes: adding random noise to the vehicle identification data, the random noise includes Gaussian noise and salt and pepper noise, the salt and pepper noise is noise that appears at random positions and has a basically fixed noise depth, and the probability density of the Gaussian noise obeys a Gaussian distribution.

[0012] In one embodiment of the present invention, the BIFPN bidirectional feature pyramid network in step S2 includes fusing features in both top-down and bottom-up directions, adding weights to each input feature to optimize the feature fusion process, and the BIFPN bidirectional feature pyramid network removes nodes with only one input edge, adds additional edges between input and output nodes at the same level, and regards each bidirectional path as a feature network layer.

[0013] In one embodiment of the present invention, the GSConv structure of the depthwise separable convolution combines the standard convolution and the depthwise separable convolution by shuffling, allowing the information of the standard convolution to be completely mixed into the output of the depthwise separable convolution by uniformly exchanging local feature information on different channels.

[0014] In one embodiment of the present invention, the use of the EIoU loss function to replace the CIoU loss function in the Yolo v5 network includes:

[0015] The EIoU loss function is to separate the influencing factors of the aspect ratio of the predicted box output by the Yolo v5 target detection algorithm and the manually calibrated real box based on the penalty term of the CIoU loss function, and calculate the length and width of the predicted box and the real box respectively. The formula of the EIoU loss function is:

[0016]

[0017] where b, b gt are the center points of the predicted box and the real box, w and h are the width and height of the predicted box, respectively. gt and h gt are the width and height of the real frame, ρ is the Euclidean distance between the two center points, and c w and c h The width and height, respectively, of the smallest bounding box that covers both boxes.

[0018] The present invention also includes a vehicle target recognition and detection system, comprising:

[0019] The annotation module is used to collect vehicle identification data, perform preprocessing on the vehicle identification data, i.e., data enhancement, to expand the data set, and create an experimental data set with the training set, validation set, and test set in the expanded data set, annotate the images of the experimental data set, and assign a label to each annotation;

[0020] An algorithm improvement module is used to train the experimental data set using the Yolo v5 target detection algorithm, improve the Yolo v5 target detection algorithm based on the single-stage detection algorithm Yolo v5 network, and introduce the BIFPN bidirectional feature pyramid network and the CBAM-nSAM attention mechanism based on the CBAM improvement;

[0021] A replacement module is used to introduce the GSConv structure based on the depth-separable convolution into the Yolo v5 target detection algorithm according to the further optimized and improved Yolo v5 target detection algorithm, replace the convolution of the Neck part in the Yolo v5 network, and use the EIoU loss function to replace the CIoU loss function in the Yolo v5 network;

[0022] The training module is used to train the improved Yolo v5 target detection algorithm using the training set and the validation set to obtain the trained model and weights, and to test the trained model and weights using the test set to evaluate the results.

[0023] The present invention also provides a vehicle target recognition and detection system, comprising:

[0024] The annotation module is used to collect vehicle identification data, perform preprocessing on the vehicle identification data, i.e., data enhancement, to expand the data set, and create an experimental data set with the training set, validation set, and test set in the expanded data set, annotate the images of the experimental data set, and assign a label to each annotation;

[0025] An algorithm improvement module is used to train the experimental data set using the Yolo v5 target detection algorithm, improve the Yolo v5 target detection algorithm based on the single-stage detection algorithm Yolo v5 network, and introduce the BIFPN bidirectional feature pyramid network and the CBAM-nSAM attention mechanism based on the CBAM improvement;

[0026] A replacement module is used to introduce the GSConv structure based on the depth-separable convolution into the Yolo v5 target detection algorithm according to the further optimized and improved Yolo v5 target detection algorithm, replace the convolution of the Neck part in the Yolo v5 network, and use the EIoU loss function to replace the CIoU loss function in the Yolo v5 network;

[0027] The training module is used to train the improved Yolo v5 target detection algorithm using the training set and the validation set to obtain the trained model and weights, and to test the trained model and weights using the test set to evaluate the results.

[0028] As described above, a vehicle target recognition and detection method and system of the present invention have the following beneficial effects:

[0029] (1) The vehicle target recognition detection method of the present invention is based on the improvement of the BIFPN bidirectional feature fusion pyramid network and the CBAM-nSAM attention mechanism, and can use the existing hardware foundation to achieve a more accurate and efficient vehicle recognition method, which can help to achieve fast and accurate vehicle recognition statistics.

[0030] (2) The vehicle target recognition detection method of the present invention uses the EIOU loss function to replace the original CIOU loss function in Yolo v5, wherein EIOU separates the influencing factors of the aspect ratio of the predicted box and the true box on the basis of the penalty term of CIOU, and calculates the length and width of the predicted box and the true box respectively, so as to solve the problem of the imbalance of difficult and easy samples in CIOU and make the position of the predicted bounding box regression more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A flowchart of a vehicle target recognition and detection method provided in an embodiment of the present application.

[0032] Figure 2 A vehicle target recognition and detection method provided in an embodiment of the present application introduces a BIFPN bidirectional feature fusion pyramid network diagram.

[0033] Figure 3 A CBAM-nSAM module diagram of a vehicle target recognition and detection method provided in an embodiment of the present application.

[0034] Figure 4 A GSConv module diagram of a vehicle target recognition and detection method provided in an embodiment of the present application.

[0035] Figure 5 An overall network structure diagram of a vehicle target recognition and detection method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0036] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0037] This section will describe in detail the specific embodiments of the present invention. The preferred embodiments of the present invention are shown in the accompanying drawings. The purpose of the accompanying drawings is to supplement the description of the text part of the specification with graphics, so that people can intuitively and vividly understand each technical feature and the overall technical solution of the present invention, but it cannot be understood as a limitation on the scope of protection of the present invention.

[0038] In the description of the present invention, it should be noted that the orientations or positional relationships indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", etc. are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention; the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be the internal communication of two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0039] It should be noted that the illustrations provided in the following embodiments are only used to illustrate the basic concept of the present invention in a schematic manner, and thus the illustrations only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0040] See also Figure 1 , Figure 1 A flow chart of a vehicle target recognition and detection method provided in an embodiment of the present application. The present invention provides a vehicle target recognition and detection method, comprising:

[0041] S1. Collect vehicle identification data and perform preprocessing, i.e., data enhancement, on the vehicle identification data to expand the data set, and use the training set, validation set, and test set in the expanded data set to create an experimental data set, annotate the images of the experimental data set, and assign a label to each annotation.

[0042] Specifically, the ratio of the training set, the validation set, and the test set is 7:1.5:1.5.

[0043] Specifically, random noise is added to the vehicle identification data, the random noise includes Gaussian noise and salt and pepper noise, the salt and pepper noise is noise that appears at random positions and has a substantially fixed noise point depth, and the probability density of the Gaussian noise obeys Gaussian distribution.

[0044] In one embodiment of the present invention, the specific steps of step S1 are as follows: first, select and combine some photos from four vehicle detection data sets, namely BIT-Vehicle, UA-DETRAC, VeRi-776 and VehicleID, and make a new data set required for the experiment according to the ratio of training set, verification set and test set of 7:1.5:1.5; then use the target detection annotation tool MAKE SENSE to mark each car on each image and assign corresponding labels to each type of car, wherein some images contain more than one label, and pay attention to ensure that the number of corresponding vehicle labels is evenly divided, and generate corresponding YOLO format label files with the suffix .txt. Preprocessing the collected vehicle identification data is to add random noise, such as Gaussian noise and salt and pepper noise, to the acquired vehicle image data. Salt and pepper noise is noise that appears at random positions and has a basically fixed noise depth. The probability density of Gaussian noise obeys the Gaussian distribution, which has two parameters, the mean value and the standard deviation. The calculation method can be obtained through the relevant knowledge about the normal distribution in probability theory, and the formula is as follows:

[0045]

[0046] Adding noise can avoid overfitting, avoid sample imbalance, and improve the robustness and generalization ability of the model.

[0047] S2. Use the Yolo v5 target detection algorithm to train the experimental data set, improve the Yolo v5 target detection algorithm based on the single-stage detection algorithm Yolo v5 network, and introduce the BIFPN bidirectional feature pyramid network and the CBAM-nSAM attention mechanism improved based on CBAM.

[0048] Specifically, the BIFPN bidirectional feature pyramid network in step S2 includes fusing features in both top-down and bottom-up directions, adding weights to each input feature to optimize the feature fusion process, and the BIFPN bidirectional feature pyramid network removes nodes with only one input edge, adds additional edges between input and output nodes at the same level, and regards each bidirectional path as a feature network layer.

[0049] In one embodiment of the present invention, in step S2, an improvement is made based on the network structure of Yolo v5, and the BIFPN network bidirectional feature fusion pyramid network is adopted to replace the original PANet structure. At the same time, a new module CBAM-nSAM is designed based on the CBAM attention module and introduced into the network structure to improve the recognition accuracy.

[0050] Please refer to Figure 2 , Figure 2 A BIFPN bidirectional feature fusion pyramid network diagram is introduced for a vehicle target recognition detection method provided in an embodiment of the present application. BIFPN adopts a weighted bidirectional fusion method, that is, it uses both top-down and bottom-up information flows for feature fusion, and adds a direct connection between the highest layer and the lowest layer to shorten the distance of information propagation; it enhances the interaction and balance between features of different scales and improves detection performance.

[0051] Please refer to Figure 3 , Figure 3 A CBAM-nSAM module diagram of a vehicle target recognition detection method provided in an embodiment of the present application. In the first half of the model, the channel attention weights are extracted by performing global maximum pooling and average pooling on the feature image. In the second half, in the spatial attention module, the channel information is temporarily ignored, and the weight results are input into multiple spatial attention modules (nSAM) to enhance the expression of local spatial information. The multi-channel results are then weighted to obtain the final feature map. The introduction of nSAM effectively improves the feature extraction capability of the network and makes small target detection more accurate.

[0052] S3. Continue to optimize the improved Yolo v5 target detection algorithm, introduce the GSConv structure based on deep separable convolution into the Yolo v5 target detection algorithm, replace the convolution of the Neck part in the Yolo v5 network, and use the EIoU loss function to replace the CIoU loss function in the Yolo v5 network.

[0053] Specifically, the EIoU loss function separates the influencing factors of the aspect ratio of the predicted box output by the Yolo v5 target detection algorithm and the manually calibrated real box based on the penalty term of the CIoU loss function, and calculates the length and width of the predicted box and the real box respectively.

[0054] The formula of the EIoU loss function is:

[0055]

[0056] where b, b gt are the center points of the predicted box and the real box, w and h are the width and height of the predicted box, respectively. gt and h gt are the width and height of the real frame, ρ is the Euclidean distance between the two center points, and c w and c h The width and height, respectively, of the smallest bounding box that covers both boxes.

[0057] In one embodiment of the present invention, a GSConv structure based on depthwise separable convolution is introduced into Yolov5 to replace the original Neck structure of the model, so as to take into account both detection accuracy and detection speed.

[0058] Please refer to Figure 4 , Figure 4 A GSConv module diagram of a vehicle target recognition detection method provided in an embodiment of the present application. GSConv is a highly efficient convolution module. GSConv first inputs a normal convolution, downsamples the normal convolution, then performs a deep convolution, and concatenates the results of the downsampled convolution and the deep convolution; finally, the corresponding channels of the previous two convolutions are shuffled together.

[0059] Assume that the size of the convolution kernel is D K *D K , the input channel is M, the output channel is N, and the output feature map size is D F *D F , the number of parameters of the standard convolution can be calculated as: D K *D K *M*N, the standard convolution calculation amount is: D K *D K *M*N*D F *D F . Depthwise separable convolution consists of depthwise convolution and pointwise convolution. Depthwise convolution is used to extract spatial features, and pointwise convolution is used to extract channel features. Depthwise separable convolution groups convolutions in the feature dimension, performs independent depthwise convolution on each channel, and uses a 1x1 convolution to aggregate all channels before output. The number of parameters for depthwise separable convolution is: D K *D K *M+M*N, the computational complexity of the depthwise separable convolution is: D K *D K *M*D F *DF +M*N*D F *D F The ratio of the parameters can be obtained as

[0060] The calculated ratio is Generally, N is large. Can be ignored, D K represents the size of the convolution kernel. If If a common 3×3 convolution kernel is used, the number of parameters and computational complexity of depthwise separable convolution is reduced to about one-ninth of the original. Compared with traditional convolution, depthwise separable convolution has fewer parameters and faster speed. We use shuffling to infiltrate the information generated by standard convolution into all parts of all information generated by depthwise separable convolution. Shuffling is a uniform mixing strategy that allows the information of standard convolution to be fully mixed into the depthwise separable output by uniformly exchanging local feature information on different channels, called GSConv. The similarity between the feature map of GSConv and standard convolution is significantly higher than the similarity between depthwise separable convolution and standard convolution. Using GSConv minimizes the negative impact of the defects of depthwise separable convolution on the model and effectively utilizes the advantages of depthwise separable convolution.

[0061] In step S3, the EIOU loss function is used to replace the original CIOU loss function in Yolo v5. EIOU separates the influencing factors of the aspect ratio of the predicted box and the true box based on the penalty term of CIOU, and calculates the length and width of the predicted box and the true box respectively to solve the problem of imbalance of difficult and easy samples in CIOU. It makes the position of the predicted bounding box regression more accurate.

[0062] Please refer to Figure 5 , Figure 5 The overall network structure diagram of a vehicle target recognition detection method provided in the embodiment of the present application. All the original Concat connection modules in Yolo v5 are replaced with BIFPN-Add, and the new CBAM-nSAM attention mechanism is placed before the output convolution. Finally, the convolution of the Neck part of the Yolo v5 target detection algorithm is replaced with GSConv. The improved overall network model is trained, the initial learning rate is set to 0.01, the input image size is 640x640, the epoch is adjusted to 300, and the batch_size is adjusted to 16.

[0063] The trained network model is evaluated using the test set and compared with the original network and other target detection networks to verify its performance. The evaluation and comparison results are as follows:

[0064]

[0065] The improved network model of the present invention improves mAP by 2.7% and mAP@0.5:0.95 by 0.5% compared with the original network, while the growth of parameter quantity and calculation amount are within an acceptable range.

[0066] S4. Use the training set and validation set to train the improved Yolo v5 target detection algorithm to obtain the trained model and weights. Use the test set to test the trained model and weights to evaluate the results.

[0067] Similar to the principle of the vehicle target recognition and detection method of the present invention, the present invention also provides a vehicle target recognition and detection system, comprising:

[0068] The annotation module is used to collect vehicle identification data, perform preprocessing on the vehicle identification data, i.e., data enhancement, to expand the data set, and create an experimental data set with the training set, validation set, and test set in the expanded data set, annotate the images of the experimental data set, and assign a label to each annotation;

[0069] An algorithm improvement module is used to train the experimental data set using the Yolo v5 target detection algorithm, improve the Yolo v5 target detection algorithm based on the single-stage detection algorithm Yolo v5 network, and introduce the BIFPN bidirectional feature pyramid network and the CBAM-nSAM attention mechanism based on the CBAM improvement;

[0070] A replacement module is used to introduce the GSConv structure based on the depth-separable convolution into the Yolo v5 target detection algorithm according to the further optimized and improved Yolo v5 target detection algorithm, replace the convolution of the Neck part in the Yolo v5 network, and use the EIoU loss function to replace the CIoU loss function in the Yolo v5 network;

[0071] The training module is used to train the improved Yolo v5 target detection algorithm using the training set and the validation set to obtain the trained model and weights, and to test the trained model and weights using the test set to evaluate the results.

[0072] In summary, the vehicle target recognition detection method of the present invention is based on the improvement of the BIFPN bidirectional feature fusion pyramid network and the CBAM-nSAM attention mechanism, and can use the existing hardware foundation to realize a more accurate and efficient vehicle recognition method, which can help to achieve fast and accurate vehicle recognition statistics. The vehicle target recognition detection method of the present invention uses the EIOU loss function to replace the original CIOU loss function in Yolo v5, wherein EIOU separates the influencing factors of the aspect ratio of the prediction box and the real box on the basis of the penalty term of CIOU, and calculates the length and width of the prediction box and the real box respectively, to solve the problem of the imbalance of difficult and easy samples in CIOU, so that the position of the predicted bounding box regression is more accurate.

[0073] The technical principle of the present invention is described above in conjunction with specific embodiments. These descriptions are only for explaining the principle of the present invention and cannot be interpreted as limiting the scope of protection of the present invention in any way. Based on the explanations herein, those skilled in the art can associate other specific embodiments of the present invention without creative work, and these equivalent variations or substitutions are all included in the scope defined by the claims of this application.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A vehicle target recognition and detection method, characterized in that: include: S1. Collect vehicle identification data, perform preprocessing on the vehicle identification data, i.e., data enhancement, to expand the data set, use the training set, validation set, and test set in the expanded data set to create an experimental data set, annotate the images of the experimental data set, and assign a label to each annotation; S2. Use the Yolo v5 target detection algorithm to train the experimental data set, improve the Yolo v5 target detection algorithm based on the single-stage detection algorithm Yolo v5 network, and introduce the BIFPN bidirectional feature pyramid network and the CBAM-nSAM attention mechanism based on the CBAM improvement; S3, continue to optimize the improved Yolo v5 target detection algorithm, introduce the GSConv structure based on deep separable convolution into the Yolo v5 target detection algorithm, replace the convolution of the Neck part in the Yolo v5 network, and use the EIoU loss function to replace the CIoU loss function in the Yolo v5 network; S4. Use the training set and validation set to train the improved Yolo v5 target detection algorithm to obtain the trained model and weights. Use the test set to test the trained model and weights to evaluate the results.

2. A vehicle target recognition and detection method according to claim 1, characterized in that: The ratio of the training set, validation set, and test set is 7:1.5:1.

5.

3. A vehicle target recognition and detection method according to claim 1, characterized in that: The data enhancement in step S1 includes: adding random noise to the vehicle identification data, the random noise includes Gaussian noise and salt and pepper noise, the salt and pepper noise is noise that appears at random positions and has a basically fixed noise depth, and the probability density of the Gaussian noise obeys Gaussian distribution.

4. A vehicle target recognition and detection method according to claim 1, characterized in that: The BIFPN bidirectional feature pyramid network in step S2 includes fusing features in both top-down and bottom-up directions, adding weights to each input feature to optimize the feature fusion process, and removing nodes with only one input edge, adding additional edges between input and output nodes at the same level, and treating each bidirectional path as a feature network layer.

5. A vehicle target recognition and detection method according to claim 1, characterized in that: The GSConv structure of the depthwise separable convolution combines the standard convolution and the depthwise separable convolution in a shuffled manner, allowing the information of the standard convolution to be completely mixed into the output of the depthwise separable convolution by uniformly exchanging local feature information on different channels.

6. A vehicle target recognition and detection method according to claim 1, characterized in that: The use of the EIoU loss function to replace the CIoU loss function in the Yolo v5 network includes: The EIoU loss function separates the influencing factors of the aspect ratio of the predicted box output by the Yolo v5 target detection algorithm and the manually calibrated real box based on the penalty term of the CIoU loss function, and calculates the length and width of the predicted box and the real box respectively. The formula of the EIoU loss function is: where b, b gt are the center points of the predicted box and the real box, w and h are the width and height of the predicted box, respectively. gt and h gt are the width and height of the real frame, ρ is the Euclidean distance between the two center points, and c w and c h The width and height, respectively, of the smallest bounding box that covers both boxes.

7. A vehicle target recognition and detection system, characterized in that: include: The annotation module is used to collect vehicle identification data, perform preprocessing on the vehicle identification data, i.e., data enhancement, to expand the data set, and create an experimental data set with the training set, validation set, and test set in the expanded data set, annotate the images of the experimental data set, and assign a label to each annotation; An algorithm improvement module is used to train the experimental data set using the Yolo v5 target detection algorithm, improve the Yolo v5 target detection algorithm based on the single-stage detection algorithm Yolo v5 network, and introduce the BIFPN bidirectional feature pyramid network and the CBAM-nSAM attention mechanism based on the CBAM improvement; A replacement module is used to introduce the GSConv structure based on the depth-separable convolution into the Yolo v5 target detection algorithm according to the further optimized and improved Yolo v5 target detection algorithm, replace the convolution of the Neck part in the Yolo v5 network, and use the EIoU loss function to replace the CIoU loss function in the Yolo v5 network; The training module is used to train the improved Yolo v5 target detection algorithm using the training set and the validation set to obtain the trained model and weights, and to test the trained model and weights using the test set to evaluate the results.

Images