Target single edge detection method based on YOLO

Abstract

The invention provides a target single-edge detection method based on YOLO, and the method comprises the steps: inputting an original image into a backbone network, carrying out the convolution processing of the original image through a multi-layer CBS module, obtaining a preliminary feature map library, inputting the preliminary feature map library into an input end of a neck structure, carrying out the multi-scale cross fusion, obtaining a final feature map library, and carrying out the Pro module of the final feature map library. And inputting the native mask, the mask coefficient and the native segmentation feature map into a mask processing module to obtain a single-edge contour image. According to the method, the YOLOV8n-seg model is set on the basis of the YOLOV8n model, and in the actual scheduling process, the detection precision of a specific single edge is improved through double models on the basis that the YOLOV8n model and the YOLOV8n-seg model are called at the same time. The double models are high in applicability, and accurate measurement data are provided on an industrial production line requiring ultrahigh precision.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer vision, and in particular relates to a target single edge detection method based on YOLO. Background Art

[0002] With the continuous advancement of industrial automation and intelligence, especially in the fields of precision robotics, automated quality inspection and manufacturing, the requirements for the accuracy and reliability of edge detection technology are constantly increasing. Although traditional edge detection algorithms, such as Canny, Sobel and Prewitt, have achieved certain success in early applications, these methods are often affected by noise and illumination changes in complex practical application scenarios, resulting in insufficient detection accuracy. To overcome these limitations, researchers have begun to adopt deep learning-based methods, such as convolutional neural networks (CNNs). These methods learn edge detection features through a large amount of training data, and can achieve more accurate and stable edge detection in a changing environment.

[0003] In the field of precision measurement, edge detection not only needs to identify the contours of objects, but also needs to complete tasks with extremely high precision, such as measuring the size of tiny parts or detecting tiny defects in products. At this time, traditional edge detection algorithms are difficult to meet high-precision requirements due to their limited recognition accuracy. Therefore, edge detection algorithms based on deep learning, such as the Holistic Edge Detection (HED) algorithm based on deep convolutional networks and the Richer Convolutional Features (RCF) algorithm based on structured learning, have been widely used to extract more detailed edge information. These algorithms can better distinguish the edges of objects in complex backgrounds, but usually require a large amount of labeled data for training.

[0004] The YOLO (You Only Look Once) model plays an important role in edge detection technology. YOLO is a real-time object detection system known for its speed and high accuracy. In the field of measurement, the YOLO model predicts object boundaries and classifications by processing the entire image at once. This one-step approach significantly improves processing speed, making it very suitable for industrial applications that require real-time measurement. After a series of version iterations, the YOLO model has continuously optimized speed and accuracy, and can use deep learning for feature extraction to more accurately identify and measure the exact location and size of objects. Especially in automated inspection lines and high-speed quality control systems, the YOLO model has begun to replace traditional image processing methods. Summary of the invention

[0005] The object of the present invention is to provide a target single edge detection method based on YOLO to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions: a single edge detection method of a target based on YOLO, a YOLOV8n-seg model set based on a YOLOV8n model, the YOLOV8n-seg model comprising a backbone network, a neck structure, a detection head, a Proto module and a mask processing module, the backbone network comprising a multi-layer CBS module, and the neck structure having a plurality of input terminals and a plurality of output terminals;

[0007] S1: The original image is input into the backbone network, and the original image is convolutionally processed by multiple layers of CBS modules. Each layer of CBS modules outputs a set of preliminary feature graphs, which are combined to obtain a preliminary feature graph library;

[0008] S2: Each group of preliminary feature map groups of the preliminary feature map library is respectively input to the input end of the neck structure, and each group of preliminary feature maps is cross-fused at multiple scales. Each output end outputs a predicted Box feature map, a predicted cls feature map and a Mask coefficient feature map, and the final feature map group is obtained by combining them. The final feature map groups output by multiple output ends are combined to obtain a final feature map library;

[0009] S3: The predicted Box feature map and the predicted cls feature map of each group in the final feature map library are input into the detection head, and the detection head is screened and processed by the classification score threshold and the NMS threshold to obtain the category information and the position information, and the target information group is obtained by combining them;

[0010] S4: The predicted Box feature map and the predicted cls feature map of each group in the final feature map library are input into the detection head to calculate the native mask and mask coefficient;

[0011] S5: The Mask coefficient feature map of each group in the final feature map library is input into the Proto module to obtain a native segmentation feature map;

[0012] S6: Input the native mask, the mask coefficient and the native segmentation feature map into a mask processing module to obtain a single edge contour image.

[0013] Preferably, the multi-layer CBS module includes a P1-CBS module, a P2-CBS module, a P3-CBS module, a P4-CBS module and a P5-CBS module, and the CBS module includes a P1-CBS module, a P2-CBS module, a P3-CBS module, a P4-CBS module and a P5-CBS module combined in series.

[0014] Preferably, the P1-CBS module is an input end, and the P3-CBS module, the P4-CBS module and the P5-CBS module are all provided with an input end P.

[0015] Preferably, the neck structure is a PAN+FPN structure, and the neck structure is provided with three input ends and three output ends.

[0016] Preferably, the S6 further comprises: obtaining final category information and position information by filtering the target information group through non-maximum suppression, and generating a predicted mask by using the native mask and the mask coefficient;

[0017] The predicted masks are batch-cropped and binarized using a threshold to obtain a single edge contour image.

[0018] Preferably, the size of the original image is 640*640, the preliminary feature map group includes multiple preliminary feature maps, and the size of the preliminary feature map is 20*20.

[0019] Preferably, the neck structure outputs three predicted Box feature maps, and the sizes of the three predicted Box feature maps are 80*80, 40*40 and 20*20 respectively.

[0020] Preferably, the neck structure outputs three predicted CLS feature maps, and the sizes of the three predicted CLS feature maps are 80*80, 40*40 and 20*20 respectively.

[0021] Preferably, the neck structure outputs three Mask coefficient feature maps, and the sizes of the three Mask coefficient feature maps are 80*80, 40*40 and 20*20 respectively.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] The present invention is based on the YOLOV8n-seg model set by the YOLOV8n model. In the actual scheduling process, the YOLOV8n model and the YOLOV8n-seg model are called simultaneously, and the accuracy of the detection of a specific single edge is improved by using the dual model. The dual model has high applicability and provides accurate measurement data on industrial production lines that require ultra-high precision.

[0024] The training cost of the present invention is low, and the dual model used requires a small training data set, and only 50 or more data sets are needed to complete the model training, and the model detection results after training are good.

[0025] The inference speed of the present invention is fast, the dual model used has superior performance, and the inference speed of a single image is about 0.11s, which meets the detection speed requirements of industrial production lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1It is a flow chart of the single edge detection method of the YOLOV8n model and the YOLOV8n-seg model of the present invention.

[0027] Figure 2 It is a flow chart of a specific single edge detection method based on YOLOV8n of the present invention.

[0028] Figure 3 It is a flowchart of a specific single edge detection method of the YOLOV8n-seg model of the present invention.

[0029] Figure 4 It is the original image to be detected in the present invention.

[0030] Figure 5 It is the detection result of Canny, the traditional edge detection algorithm.

[0031] Figure 6 It is the detection result of the deep learning LDC edge detection model.

[0032] Figure 7 It is the detection result of the deep learning HED edge detection model.

[0033] Figure 8 The present invention detects results based on a specific single edge detection method of the YOLOV8n model. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of this protection.

[0035] Embodiment 1:

[0036] like Figure 1-Figure 8As shown, a target single edge detection method based on YOLO provided by the present invention is a YOLOV8n-seg model set based on the YOLOV8n model, wherein the YOLOV8n-seg model includes a backbone network, a neck structure, a detection head, a Proto module and a mask processing module, the backbone network includes a multi-layer CBS module, and the neck structure is provided with multiple input terminals and multiple output terminals, S1: the original image is input into the backbone network, the original image is convolved through the multi-layer CBS module, and each layer of the CBS module outputs a group of preliminary feature map groups, which are combined to obtain a preliminary feature map library; S2: each group of preliminary feature map groups in the preliminary feature map library is respectively input into the input terminal of the neck structure, and each group of preliminary feature maps is cross-fused at multiple scales, and each output terminal outputs a predicted Box feature map, a predicted The cls feature map and the Mask coefficient feature map are combined to obtain the final feature map group, and the final feature map groups output by multiple output terminals are combined to obtain the final feature map library; S3: the predicted Box feature map and the predicted cls feature map of each group in the final feature map library are input into the detection head, and the detection head is screened and processed by the classification score threshold and the NMS threshold to obtain the category information and position information, and the target information group is combined; S4: the predicted Box feature map and the predicted cls feature map of each group in the final feature map library are input into the detection head, and the native mask and mask coefficient are calculated; S5: the Mask coefficient feature map of each group in the final feature map library is input into the Proto module to obtain the native segmentation feature map; S6: the native mask, mask coefficient and native segmentation feature map are input into the mask processing module to obtain a single edge contour image. The multi-layer CBS module includes a P1-CBS module, a P2-CBS module, a P3-CBS module, a P4-CBS module and a P5-CBS module, and the CBS module includes a P1-CBS module, a P2-CBS module, a P3-CBS module, a P4-CBS module and a P5-CBS module, which are combined in series. The P1-CBS module is the input end, and the P3-CBS module, the P4-CBS module and the P5-CBS module are all provided with an input end P. The neck structure is a PAN+FPN structure, and the neck structure is provided with 3 input ends and 3 output ends. S6 also includes: the target information group is screened by non-maximum suppression to obtain the final category information and position information, and the prediction mask is generated using the native mask and the mask coefficient; the prediction mask is batch cropped, and the threshold is used for binarization to obtain a single edge contour image. The size of the original image is 640*640, and the preliminary feature map group includes multiple preliminary feature maps, and the size of the preliminary feature map is 20*20. The neck structure outputs 3 predicted Box feature maps, and the sizes of the 3 predicted Box feature maps are 80*80, 40*40, and 20*20 respectively. The neck structure outputs 3 predicted CLS feature maps, and the sizes of the 3 predicted CLS feature maps are 80*80, 40*40, and 20*20 respectively.The neck structure outputs three Mask coefficient feature maps, and the sizes of the three Mask coefficient feature maps are 80*80, 40*40 and 20*20 respectively.

[0037] Through the above technical solution, the present invention sets a YOLOV8n-seg model based on the YOLOV8n model. In the actual scheduling process, the YOLOV8n model and the YOLOV8n-seg model are called simultaneously, and the accuracy of the detection of a specific single edge is improved through the dual model. The dual model has high applicability and provides accurate measurement data on industrial production lines that require ultra-high precision.

[0038] The training cost of the present invention is low, and the dual model used requires a small training data set, and only 50 or more data sets are needed to complete the model training, and the model detection results after training are good.

[0039] The inference speed of the present invention is fast, the dual model used has superior performance, and the inference speed of a single image is about 0.11s, which meets the detection speed requirements of industrial production lines.

[0040] Embodiment 2:

[0041] YOLO (You Only Look Once) is an object detection algorithm based on deep learning. It has achieved remarkable results in the field of object detection with its high speed and high precision. The core idea of ​​the YOLO model is to transform the object detection task into a regression problem, and predict the bounding box and category of the object by processing the entire image at one time. This method not only improves the detection speed, but also reduces the computational complexity, making the YOLO model very suitable for real-time detection tasks.

[0042] like Figure 1-Figure 8 As shown, for the single edge detection task, the present invention proposes a specific single edge automatic detection method based on YOLO. This method combines the YOLOV8n and YOLOV8n-seg models, and realizes accurate detection of specific single edges by fusing the two models.

[0043] Data annotation and preprocessing:

[0044] First, you need to use the public labeling software Labelme to label the edge image to be detected. During the labeling process, you need to label two areas separately:

[0045] The region where the edge to be detected is located (region): This is a larger region that contains the location of the edge to be detected.

[0046] The edge of the object to be detected: This is an accurate outline that indicates the edge position of the object.

[0047] After the annotation is completed, the generated annotation file needs to be preprocessed. The preprocessing steps include data format conversion, normalization, etc. to ensure that the data can be correctly read and processed by the YOLOV8n and YOLOV8n-seg models.

[0048] YOLOV8n model training:

[0049] The YOLOV8n model is used to generate a mask image of the edge area to be detected. The model training process includes the following steps:

[0050] Model construction: The backbone network of the YOLOV8n model consists of multiple CBS modules, C2f modules, and SPPF modules. The CBS module is used to perform convolution processing on the input image to extract features. The C2f module and SPPF module further process and optimize the feature map.

[0051] Training data preparation: Input the labeled data into the YOLOV8n model for training. The training process includes steps such as forward propagation, loss calculation, back propagation, and optimization update.

[0052] Model training: Use the training data to train the YOLOV8n model and optimize the model parameters to obtain the best edge area mask prediction results.

[0053] YOLOV8n-seg model training:

[0054] The YOLOV8n-seg model is used to generate a mask image of the object contour where the edge to be detected is located. The model training process is similar to YOLOV8n, but there are some differences in the detection head (head layer):

[0055] Detection head design: The YOLOV8n-seg model generates additional feature maps in the detection head layer. In addition to generating feature maps for Box prediction and cls prediction, feature maps for mask coefficients (mask_coefficients) are also generated.

[0056] Prototype Mask generation: A Prototype Mask feature map of size (1, 32, 80, 80) is generated through a feature map of 80×80 scale to generate a feature map for native segmentation.

[0057] Model training: Input the labeled data into the YOLOV8n-seg model for training and optimize the model parameters to obtain the best edge contour mask prediction results.

[0058] Post-processing and result fusion:

[0059] After training, the model generates region masks and edge masks. The post-processing steps include:

[0060] Mask fusion: Fuse the region mask generated by the YOLOV8n model and the edge mask generated by the YOLOV8n-seg model, and take the intersection image to obtain the final single edge detection result.

[0061] Mask cropping and threshold processing: For the mask generated by the YOLOV8n-seg model, cropping and threshold processing are performed to obtain the final mask outline result.

[0062] This method has the following advantages in single edge detection tasks:

[0063] High-precision automatic detection:

[0064] By using the YOLOV8n and YOLOV8n-seg models for dual model fusion, this method can achieve intelligent automatic detection in ultra-high precision scenes. The YOLOV8n model is responsible for locating the edge area, while the YOLOV8n-seg model is responsible for generating accurate edge contour masks. This dual processing makes the detection results more accurate and stable.

[0065] Dual-model hybrid detection:

[0066] This method proposes the idea of ​​dual-model hybrid detection. By combining the YOLOV8n and YOLOV8n-seg models, it can effectively solve the noise and interference problems in traditional single edge detection methods. The YOLOV8n model locates the edge area, while the YOLOV8n-seg model improves the detection accuracy by generating detailed masks.

[0067] Efficient mask generation and fusion:

[0068] The YOLOV8n-seg model can provide more detailed mask information in target detection by generating additional mask coefficients and Prototype Mask feature maps. Combined with the edge area mask of the YOLOV8n model, more accurate mask fusion can be achieved, thereby improving the accuracy of single edge detection.

[0069] Superior to traditional algorithms and models:

[0070] Compared with traditional edge detection algorithms (such as Canny, Sobel) and common deep learning edge detection models (such as LDC, HED), this method performs better in specific single edge automatic detection tasks. Traditional methods usually perform poorly when dealing with complex backgrounds and noise, while this method uses the powerful feature extraction capabilities of deep learning models to effectively reduce noise interference and improve detection accuracy.

[0071] In order to verify the reliability of the single edge detection method of the present invention, a comparative experiment was carried out to compare three edge detection algorithms: the traditional edge detection algorithm Canny, the deep learning LDC edge detection model, and the deep learning HED edge detection model.

[0072] Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 They are respectively the original image to be detected, the detection result of the traditional edge detection algorithm Canny, the detection result of the deep learning LDC edge detection model, the detection result of the deep learning HED edge detection model, and the detection result of the specific single edge detection method based on YOLO of the present invention.

[0073] The detection tasks of traditional edge detection algorithms and common deep learning edge detection models such as the Canny algorithm, LDC model, and HED model are all aimed at full-image detection and do not meet the requirements of single-edge automatic detection tasks. Even if the present invention provides auxiliary positioning using the YOLOV8n model or other target detection models, a lot of noise and interference will be generated in the area of ​​the edge to be detected. In terms of performance, the present invention is significantly superior to the above algorithms and models. The specific single-edge automatic detection method based on YOLO has great potential.

[0074] Embodiment three:

[0075] like Figure 1 As shown, the specific single edge detection method based on YOLO provided in this embodiment includes the following steps:

[0076] Use the public annotation software Lableme to annotate the edge image, and mark the region where the edge to be detected is located and the outline of the object where the edge to be detected is located.

[0077] The obtained annotation files are preprocessed and sent to the YOLOV8n and YOLOV8n-seg models for training respectively, and the region mask image of the edge area to be detected and the edge mask image of the object contour where the edge to be detected is located are inferred respectively.

[0078] The region mask and edge mask are post-processed, and the intersection image is fused to finally obtain a single-edge detection result containing only the edge to be detected.

[0079] like Figure 2As shown in the figure, the backbone network extracts features from the input image and is mainly composed of CBS modules, C2f modules and SPPF modules. The CBS module represents Conv, BN and SILU, which are generally used to convolve the input image or feature map. The result of convolution is a decrease in resolution and an increase in the number of channels. There are a total of 5 CBS modules in the backbone network. The input image resolution changes from 640*640 to 20*20, and the number of input channels changes from 3 to 512.

[0080] The neck structure is used to fuse the feature maps output by the backbone network, that is, to concatenate the features extracted from the three different scales through upsampling. After two upsamplings, the first feature map (80*80) is output to the head layer, and then the feature map is downsampled through the CBS module to output the remaining two feature maps (40*40 and 20*20) to the head layer.

[0081] The detection head layer network is designed to generate specific feature maps based on the number of categories. YOLOV8 uses a decoupled head to generate feature maps. Decoupling means generating feature maps for predicting cls and Box respectively. For example, the input scale is (1, 256, 80, 80) feature map P3. The branch that generates the Box prediction feature map will generate a feature map of (1, nc, 80, 80) through two CBS modules and a Cov2d. nc represents the predicted category. After the head layer, three feature maps for predicting Box (1, 64, 80, 80), (1, 64, 40, 40) and (1, 64, 20, 20) will be generated; and three feature maps for predicting cls (1, nc, 80, 80), (1, nc, 40, 40) and (1, nc, 20, 20) will be generated. Finally, the feature map is converted into two forms: (1, 8400, nc) and (1, 8400, 4). After being screened by some classification score thresholds and NMS thresholds, the category information and location information of the target are finally obtained.

[0082] like Figure 3 As shown in the figure, in the YOLOV8n-seg model of instance segmentation, the backbone network, neck structure and target detection YOLOV8n model are exactly the same, except that the detection head layer is different. There are two differences: First, the feature maps of the three scales are generated in the head layer in addition to generating three feature maps of predicted Box (1, 64, 80, 80), (1, 64, 40, 40) and (1, 64, 20, 20); and three feature maps of predicted cl s (1, nc, 80, 80), (1, nc, 40, 40) and (1, nc, 20, 20); 3 feature maps with 32 channels are also generated in the head layer to be used as Mask coefficients. _coefficients(1, 32, 80, 80), (1, 32, 40, 40) and (1, 32, 20, 20); secondly, a Prototype Mask feature map of size (1, 32, 80, 80) is generated through the 80*80 scale feature map as the feature map of native segmentation. Finally, two targets are selected through NMS, and the predicted masks of the two targets are generated with the help of the native mask and mask coefficients. Then, the predicted mask is cropped according to the predicted box, and then binarized using the Threshold threshold to obtain the final mask contour result.

[0083] Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 They are respectively the original image to be detected, the detection result of the traditional edge detection algorithm Canny, the detection result of the deep learning LDC edge detection model, the detection result of the deep learning HED edge detection model, and the detection result of the specific single edge detection method based on YOLO of the present invention.

[0084] The detection tasks of traditional edge detection algorithms and common deep learning edge detection models such as the Canny algorithm, LDC model, and HED model are all aimed at full-image detection and do not meet the task of single-edge automatic detection. Even if the present invention provides auxiliary positioning using the YOLOV8n model or other target detection models, a lot of noise and interference will be generated in the area of ​​the edge to be detected.

[0085] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article 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, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0086] The above description is only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in this field should be included in the scope of the claims as long as they do not depart from the spirit and scope of the technical solution of the present invention.

Claims

1. A single edge detection method based on YOLO, It is characterized in that A YOLOV8n-seg model based on a YOLOV8n model, wherein the YOLOV8n-seg model includes a backbone network, a neck structure, a detection head, a Proto module, and a mask processing module, wherein the backbone network includes a multi-layer CBS module, and the neck structure is provided with a plurality of input terminals and a plurality of output terminals; S1: The original image is input into the backbone network, and the original image is convolutionally processed by multiple layers of CBS modules. Each layer of CBS modules outputs a set of preliminary feature graphs, which are combined to obtain a preliminary feature graph library; S2: Each group of preliminary feature map groups of the preliminary feature map library is respectively input to the input end of the neck structure, and each group of preliminary feature maps is cross-fused at multiple scales. Each output end outputs a predicted Box feature map, a predicted cls feature map and a Mask coefficient feature map, and the final feature map group is obtained by combining them. The final feature map groups output by multiple output ends are combined to obtain a final feature map library; S3: The predicted Box feature map and the predicted cls feature map of each group in the final feature map library are input into the detection head, and the detection head is screened and processed by the classification score threshold and the NMS threshold to obtain the category information and the position information, and the target information group is obtained by combining them; S4: The predicted Box feature map and the predicted cls feature map of each group in the final feature map library are input into the detection head to calculate the native mask and mask coefficient; S5: The Mask coefficient feature map of each group in the final feature map library is input into the Proto module to obtain a native segmentation feature map; S6: Input the native mask, the mask coefficient and the native segmentation feature map into a mask processing module to obtain a single edge contour image.

2. A target single edge detection method based on YOLO according to claim 1, It is characterized in that The multi-layer CBS module includes a P1-CBS module, a P2-CBS module, a P3-CBS module, a P4-CBS module and a P5-CBS module, and the CBS module includes a P1-CBS module, a P2-CBS module, a P3-CBS module, a P4-CBS module and a P5-CBS module combined in series.

3. According to the YOL0-based target single edge detection method of claim 2, It is characterized in that The P1-CBS module is an input terminal, and the P3-CBS module, the P4-CBS module and the P5-CBS module are all provided with an input terminal P.

4. A target single edge detection method based on YOLO according to claim 1, It is characterized in that The neck structure is a PAN+FPN structure, and the neck structure is provided with three input ends and three output ends.

5. A target single edge detection method based on YOLO according to claim 1, It is characterized in that The step S6 further includes: filtering the target information group through non-maximum suppression to obtain final category information and position information, and generating a predicted mask using the native mask and the mask coefficient; The predicted masks are batch-cropped and binarized using a threshold to obtain a single edge contour image.

6. A target single edge detection method based on YOLO according to claim 1, It is characterized in that The size of the original image is 640*640, the preliminary feature map group includes multiple preliminary feature maps, and the size of the preliminary feature map is 20*20.

7. The target single edge detection method based on YOLO according to claim 1, It is characterized in that The neck structure outputs three predicted Box feature maps, and the sizes of the three predicted Box feature maps are 80*80, 40*40 and 20*20 respectively.

8. The target single edge detection method based on YOL0 according to claim 1, It is characterized in that The neck structure outputs three predicted CLS feature maps, and the sizes of the three predicted CLS feature maps are 80*80, 40*40 and 20*20 respectively.

9. The target single edge detection method based on YOLO according to claim 1, It is characterized in that The neck structure outputs three Mask coefficient feature maps, and the sizes of the three Mask coefficient feature maps are 80*80, 40*40 and 20*20 respectively.

Images