A sperm identification method and system combining microfluidic chip and machine vision

Abstract

The present invention discloses a sperm recognition method and system combining a microfluidic chip with machine vision, and relates to the technical field of image recognition. By placing a sperm sample in a microfluidic chip, after the sperm settles, an image of each micropit in the microfluidic chip is obtained, and encoding is performed so that each image corresponds to a horizontal and vertical coordinate position; the target image is used as the input of a pre-trained target recognition model to obtain the recognition result of the target image; the recognition result includes whether there is normal sperm and the sperm position; the target image is the image of any micropit; according to the recognition result of each image, the location and number of normal sperm on the microfluidic chip are determined. Through the pre-trained target recognition model, it is possible to automatically identify whether there is sperm and sperm morphology in the image, so that normal sperm can be efficiently screened out, and sperm positioning information is provided for subsequent capture operations.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a sperm recognition method and system combining a microfluidic chip with machine vision. Background Art

[0002] In the field of modern reproductive medicine, solving the problem of male infertility has always been one of the research focuses. Among them, oligospermia, especially severe oligospermia, is currently one of the most common causes of male infertility. The number of sperm in the semen of patients with severe oligospermia is extremely small. It is difficult and time-consuming to find sperm when using conventional gradient centrifugation technology to process the semen of patients with severe oligospermia. In addition, sperm is easily damaged during the gradient centrifugation process, so that it is difficult for patients with severe oligospermia to find a sufficient number of normal sperm for assisted reproduction.

[0003] In order to solve this problem, the application of microfluidic chip technology in the biomedical field has gradually attracted attention in recent years. Microfluidic chip technology has shown great potential in sperm capture and screening with its advantages of integrated miniaturization and high-throughput rapid detection. However, how to efficiently and accurately identify and locate sperm in the semen of patients with severe oligospermia on a microfluidic chip is still a problem that needs to be solved. Although the traditional microscopic observation method can be used to identify sperm, manual identification has a low degree of automation, low efficiency, long time consumption, and a high missed detection rate. Summary of the invention

[0004] The purpose of the present invention is to solve the problems of low automation and low efficiency mentioned in the above background technology, and to propose a sperm identification method and system combining a microfluidic chip with machine vision.

[0005] In a first aspect of the present invention, a sperm identification method combining a microfluidic chip and machine vision is provided, wherein the microfluidic chip has a high-density micro-pit array; the method comprises:

[0006] Place the sperm sample in a microfluidic chip;

[0007] After the sperm settles, an image of each micropit in the microfluidic chip is obtained and encoded so that each image corresponds to a position of a horizontal and vertical coordinate;

[0008] The target image is used as an input of a pre-trained target recognition model to obtain a recognition result of the target image; the recognition result includes whether there is normal sperm and the position of the sperm; the target image is an image of any micro-pit;

[0009] Based on the recognition results of each image, the location and number of normal sperm on the microfluidic chip are determined.

[0010] Optionally, the target recognition model is improved based on the YOLOv8 model, and the specific improvements include:

[0011] Keep the first CBS module in the backbone structure and replace the remaining CBS modules with SPDConv modules;

[0012] Before the SPPF module, a multi-scale feature enhancement module is inserted;

[0013] An attention module is added between the jump connections of the trunk structure and the neck structure, combined with upsampling and channel concatenation to form a feature fusion module;

[0014] Add a small object detection layer to the head structure.

[0015] Optionally, the multi-scale feature enhancement module includes multiple deep convolution branches; and the calculation process of the multi-scale feature enhancement module includes:

[0016] ;

[0017] Among them, X0 is the input of the multi-scale feature enhancement module, X2, X3 and X4 are the outputs of the three deep convolution branches respectively; W is the attention weight; Y is the output of the multi-scale feature enhancement module; f is the operation symbol, the subscripts Conv and DWConv represent standard convolution and deep convolution respectively, and the superscript represents the convolution kernel size.

[0018] Optionally, the calculation process of the feature fusion module includes:

[0019] Receive the feature map transmitted by the backbone structure and perform spatial attention weighting on it to obtain the first feature map;

[0020] receiving a feature map transmitted by the neck structure and upsampling the feature map to obtain a second feature map;

[0021] The first feature map and the second feature map are concatenated in the channel dimension to obtain and output the third feature map.

[0022] Optionally, during the model training process, an EIoU loss function is used as the bounding box regression loss of the detection head; the calculation formula of the EIoU loss function is:

[0023] ;

[0024] Among them, L EIoU is the desired EIoU loss; L IoU is the IoU loss, IoU is the intersection over union ratio of the predicted box and the real box; L dis is the distance loss, dc is the distance between the center of the predicted box and the center of the real box, W p and Hp L is the width and height of the minimum bounding box of the predicted box and the real box respectively; asp is the edge length loss, dw and dh are the width difference and height difference between the predicted box and the real box respectively.

[0025] A second aspect of the present invention provides a sperm recognition system combining a microfluidic chip with machine vision, the system comprising:

[0026] A sample acquisition module, used to place the sperm sample in the microfluidic chip;

[0027] An image acquisition module is used to acquire the image of each micropit in the microfluidic chip after the sperm has settled, and encode it so that each image corresponds to a position of a horizontal and vertical coordinate;

[0028] The sperm recognition module is used to use the target image as the input of the pre-trained target recognition model to obtain the recognition result of the target image; the recognition result includes whether there are normal sperm and the sperm position; the target image is the image of any micro-pit;

[0029] The information integration module is used to determine the location and quantity of normal sperm on the microfluidic chip based on the recognition results of each image.

[0030] Optionally, the target recognition model is improved based on the YOLOv8 model, and the specific improvements include:

[0031] Keep the first CBS module in the backbone structure and replace the remaining CBS modules with SPDConv modules;

[0032] Before the SPPF module, a multi-scale feature enhancement module is inserted;

[0033] An attention module is added between the jump connections of the trunk structure and the neck structure, combined with upsampling and channel concatenation to form a feature fusion module;

[0034] Add a small object detection layer to the head structure.

[0035] Optionally, the multi-scale feature enhancement module includes multiple deep convolution branches; and the calculation process of the multi-scale feature enhancement module includes:

[0036] ;

[0037] Among them, X0 is the input of the multi-scale feature enhancement module, X2, X3 and X4 are the outputs of the three deep convolution branches respectively; W is the attention weight; Y is the output of the multi-scale feature enhancement module; f is the operation symbol, the subscripts Conv and DWConv represent standard convolution and deep convolution respectively, and the superscript represents the convolution kernel size.

[0038] Optionally, the calculation process of the feature fusion module includes:

[0039] Receive the feature map transmitted by the backbone structure and perform spatial attention weighting on it to obtain the first feature map;

[0040] receiving a feature map transmitted by the neck structure and upsampling the feature map to obtain a second feature map;

[0041] The first feature map and the second feature map are concatenated in the channel dimension to obtain and output the third feature map.

[0042] Optionally, during the model training process, an EIoU loss function is used as the bounding box regression loss of the detection head; the calculation formula of the EIoU loss function is:

[0043] ;

[0044] Among them, L EIoU is the desired EIoU loss; L IoU is the IoU loss, IoU is the intersection over union ratio of the predicted box and the real box; L dis is the distance loss, dc is the distance between the center of the predicted box and the center of the real box, W p and H p L is the width and height of the minimum bounding box of the predicted box and the real box respectively; asp is the edge length loss, dw and dh are the width difference and height difference between the predicted box and the real box respectively.

[0045] Beneficial effects of the present invention:

[0046] The present invention proposes a sperm recognition method combining a microfluidic chip with machine vision, wherein the microfluidic chip has a high-density micropit array; the method comprises: placing a sperm sample in the microfluidic chip; after the sperm settles, obtaining an image of each micropit in the microfluidic chip, and encoding the image so that each image corresponds to a position of a horizontal and vertical coordinate; using a target image as an input of a pre-trained target recognition model to obtain a recognition result of the target image; the recognition result includes whether there is normal sperm and the position of the sperm; the target image is an image of any micropit; and according to the recognition result of each image, determining the position and quantity of normal sperm on the microfluidic chip.

[0047] Through the pre-trained target recognition model, the presence or absence of sperm and sperm morphology in the image can be automatically identified, so that normal sperm can be efficiently screened out and sperm positioning information can be provided for subsequent capture operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The present invention will be further described below in conjunction with the accompanying drawings.

[0049] Figure 1 A flowchart of a sperm identification method combining a microfluidic chip and machine vision is provided for an embodiment of the present invention;

[0050] Figure 2 A schematic diagram of a micro-pit array in a chip after sperm sedimentation is provided for an embodiment of the present invention;

[0051] Figure 3 A network structure diagram of a YOLOv8 model is provided for an embodiment of the present invention;

[0052] Figure 4 A network structure diagram of a target recognition model is provided for an embodiment of the present invention;

[0053] Figure 5 A schematic structural diagram of a multi-scale feature enhancement module is provided for an embodiment of the present invention;

[0054] Figure 6 A structural schematic diagram of a feature fusion module is provided for an embodiment of the present invention;

[0055] Figure 7 A schematic diagram of a recognition result is provided for an embodiment of the present invention;

[0056] Figure 8 An architectural diagram of a sperm recognition system combining a microfluidic chip and machine vision is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0057] 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 of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] The embodiment of the present invention provides a sperm identification method combining a microfluidic chip with machine vision, wherein the microfluidic chip has a high-density micro-pit array. Figure 1 , Figure 1 A flowchart of a sperm identification method combining a microfluidic chip and machine vision provided in an embodiment of the present invention. The method comprises the following steps:

[0059] S101, placing the sperm sample in a microfluidic chip.

[0060] S102, after the sperm settles, an image of each micropit in the microfluidic chip is obtained and encoded so that each image corresponds to a position of a horizontal and vertical coordinate.

[0061] S103, using the target image as an input of a pre-trained target recognition model to obtain a recognition result of the target image.

[0062] S104, determining the location and quantity of normal sperm on the microfluidic chip according to the recognition result of each image.

[0063] The recognition result includes whether there are normal sperm and the sperm position; the target image is the image of any micro-pit.

[0064] A sperm recognition method combining a microfluidic chip and machine vision provided in an embodiment of the present invention can automatically identify the presence or absence of sperm and the sperm morphology in an image through a pre-trained target recognition model, thereby efficiently screening out normal sperm and providing sperm positioning information for subsequent capture operations.

[0065] In one implementation, the sperm sample is diluted and placed in the chip. After 10-20 minutes of natural sedimentation, the sperm can enter the micro-pit array of the chip. After sedimentation, the micro-pits in the chip are photographed using an optical microscope to obtain a micro-pit image. Figure 2 , Figure 2 A schematic diagram of a micro-pit array in a chip after sperm sedimentation provided by an embodiment of the present invention. The shape of the micro-pit is not limited to the hexagonal shape shown in the figure, but can also be rectangular, circular, etc.

[0066] In one implementation, historical sperm sample images are collected and annotated to obtain a data set. The annotation types include normal sperm, head deformity, mid-section abnormality, tail abnormality, etc. The data set is used for model training to obtain a target recognition model. After one recognition operation, the location and number of normal sperm are obtained, and a table file is generated, so that the micromanipulation system can be used to accurately capture and collect sperm in the micropit.

[0067] In one embodiment, the target recognition model is improved based on the YOLOv8 model. Figure 3 , Figure 3 A network structure diagram of a YOLOv8 model provided in an embodiment of the present invention. The YOLOv8 model includes a backbone structure Backbone, a neck structure Neck, and a head structure Head; the backbone structure is used for feature extraction, the neck structure is used for feature fusion, and the head structure is used for target recognition and classification. Specific improvements include:

[0068] Keep the first CBS module in the backbone structure and replace the remaining CBS modules with SPDConv modules;

[0069] Before the SPPF module, a multi-scale feature enhancement module is inserted;

[0070] An attention module is added between the jump connections of the trunk structure and the neck structure, combined with upsampling and channel concatenation to form a feature fusion module;

[0071] Add a small object detection layer to the head structure.

[0072] See also Figure 4 , Figure 4 A network structure diagram of a target recognition model provided by an embodiment of the present invention. The target recognition model includes a backbone structure, a neck structure, and a head structure; the backbone structure includes a first CBS module (CBS_1), a first SPDConv module (SPDConv_1), a first C2f module (C2f_1), a second SPDConv module (SPDConv_2), a second C2f module (C2f_2), a third SPDConv module (SPDConv_3), a third C2f module (C2f_3), a fourth SPDConv module (SPDConv_4), and a fourth C2f module (C2f_4) connected in sequence. ), multi-scale feature enhancement module (FEM) and SPPF module (SPPF); the neck structure includes the first feature fusion module (FFM_1), the fifth C2f module (C2f_5), the second feature fusion module (FFM_2), the sixth C2f module (C2f_6), the third feature fusion module (FFM_3), the seventh C2f module (C2f_7), the fifth SPDConv module (SPDConv_5), the first splicing module (Concat_1), the eighth C2f module (C2f_8), and the sixth SPDConv module connected in sequence. (SPDConv_6), the second splicing module (Concat_2), the ninth C2f module (C2f_9), the seventh SPDConv module (SPDConv_7), the third splicing module (Concat_3) and the tenth C2f module (C2f_10); the head structure includes a first detection head (detect1), a second detection head (detect2), a third detection head (detect3) and a fourth detection head (detect4); the first detection head receives the output of the tenth C2f module as input; the second detection head receives the output of the tenth C2f module as input; the third detection head receives the output of the ninth C2f module as input; the fourth detection head receives the output of the eighth C2f module as input; wherein, the first fusion module also receives the output of the third C2f module as input; the second fusion module also receives the output of the second C2f module as input; the third fusion module also receives the output of the first C2f module as input; the first splicing module also receives the output of the sixth C2f module as input; the second splicing module also receives the output of the fifth C2f module as input; the third splicing module also receives the output of the SPPF module as input.

[0073] In one implementation, the SPDConv (Space-to-Depth Convolution) module consists of a space-to-depth layer (SPD) and a non-stride convolution layer (Conv). SPD is used to divide the spatial dimension of the input feature map into different blocks (reducing the spatial dimension to half of the original), and the pixel values ​​in each block are rearranged into different depth channels (expanding the channel dimension to four times the original). The non-stride convolution layer is then used for feature extraction, and a 3×3 convolution kernel can be used. This space-to-depth conversion can increase the network's receptive field, allowing the network to pay more attention to fine-grained features.

[0074] In one implementation, see Figure 5 , Figure 5 A schematic diagram of the structure of a multi-scale feature enhancement module provided by an embodiment of the present invention. As shown in the figure, the multi-scale feature enhancement module includes multiple deep convolution branches; the calculation process of the multi-scale feature enhancement module includes:

[0075] ;

[0076] Among them, X0 is the input of the multi-scale feature enhancement module, X2, X3 and X4 are the outputs of the three deep convolution branches respectively; W is the attention weight; Y is the output of the multi-scale feature enhancement module; f is the operator symbol, the subscripts Conv and DWConv represent standard convolution and deep convolution respectively, and the superscript represents the convolution kernel size; the symbol Represents element-wise multiplication.

[0077] This feature enhancement module can effectively handle targets of different sizes by aggregating multi-scale features. For sperm recognition tasks, it can capture both large background information and tiny details of sperm. This module is inserted at the bottom of the backbone structure. By connecting this module after multiple convolutional layers, it can refine the features of sperm targets before further deep feature extraction, thereby ensuring recognition accuracy. A two-dimensional convolution kernel (such as 7×7) is decomposed into two one-dimensional convolution kernels (such as 1×7 and 7×1), thereby reducing computational complexity.

[0078] In one implementation, see Figure 6 , Figure 6 A structural diagram of a feature fusion module provided by an embodiment of the present invention. In the figure, LP is a feature map from the trunk structure; HP is a feature map from the neck structure. The calculation process of the feature fusion module includes:

[0079] Receive the feature map transmitted by the backbone structure and perform spatial attention weighting on it to obtain the first feature map;

[0080] receiving a feature map transmitted by the neck structure and upsampling the feature map to obtain a second feature map;

[0081] The first feature map and the second feature map are concatenated in the channel dimension to obtain and output the third feature map.

[0082] like Figure 6 As shown in the upper middle part, the spatial attention weighting includes: pooling (maximum pooling GMP and average pooling GAP in the channel dimension respectively); concatenating the pooling results and inputting them into a shared convolutional layer, and generating a spatial attention map through the Sigmoid activation function; using the spatial attention map to weight the input feature map LP to obtain the first feature map. By performing spatial attention weighting on the feature map transmitted by the backbone structure, the key areas in the feature map that are useful for target detection or recognition (such as the sperm head or tail) can be highlighted, while suppressing background noise.

[0083] Upsampling can be achieved through linear interpolation. Upsampling the neck feature map can improve its spatial resolution and align it with the feature map of the backbone structure in spatial size for easy fusion.

[0084] In one embodiment, during the model training process, the EIoU loss function is used as the bounding box regression loss of the detection head; the calculation formula of the EIoU loss function is:

[0085] ;

[0086] Among them, L EIoU is the desired EIoU loss; L IoU is the IoU loss, IoU is the intersection over union ratio of the predicted box and the real box; L dis is the distance loss, dc is the distance between the center of the predicted box and the center of the real box, W p and H p L is the width and height of the minimum bounding box of the predicted box and the real box respectively; asp is the edge length loss, dw and dh are the width difference and height difference between the predicted box and the real box respectively.

[0087] In one implementation, the bounding box regression loss of the original YOLOv8 model uses CIoU. Both EIoU and CIoU are variants of enhanced IoU loss, but after adding distance loss and edge length loss, EIoU can still provide effective gradients to guide model optimization even if the two bounding boxes do not overlap. This means that EIoU can provide more stable gradient updates in the early stages of training or when the target box prediction is poor. Compared with CIoU, it can better avoid gradient vanishing or gradient instability, thereby promoting model training and convergence.

[0088] See also Figure 7, Figure 7 A schematic diagram of a recognition result provided by an embodiment of the present invention. Figure 7 The prediction box of normal sperm and its confidence are shown, and the number of sperm can be counted based on the recognition results.

[0089] The embodiment of the present invention provides a sperm recognition system combining a microfluidic chip with machine vision, wherein the microfluidic chip has a high-density micro-pit array. Figure 8 , Figure 8 The architecture diagram of a sperm recognition system combining a microfluidic chip and machine vision provided in an embodiment of the present invention. The system includes:

[0090] The sample acquisition module is used to place the sperm sample in the microfluidic chip.

[0091] The image acquisition module is used to acquire the image of each micropit in the microfluidic chip after the sperm settles, and encode it so that each image corresponds to a horizontal and vertical coordinate position.

[0092] The sperm recognition module is used to take the target image as the input of the pre-trained target recognition model to obtain the recognition result of the target image.

[0093] The information integration module is used to determine the location and quantity of normal sperm on the microfluidic chip based on the recognition results of each image.

[0094] The recognition result includes whether there are normal sperm and the sperm position; the target image is the image of any micro-pit.

[0095] A sperm recognition system combining a microfluidic chip and machine vision provided in an embodiment of the present invention can automatically identify the presence or absence of sperm and the sperm morphology in an image through a pre-trained target recognition model, thereby efficiently screening out normal sperm and providing sperm positioning information for subsequent capture operations.

[0096] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A sperm identification method combining a microfluidic chip and machine vision, characterized in that: The microfluidic chip has a high-density micro-pit array; the method comprises: Place the sperm sample in a microfluidic chip; After the sperm settles, an image of each micropit in the microfluidic chip is obtained and encoded so that each image corresponds to a position of a horizontal and vertical coordinate; The target image is used as an input of a pre-trained target recognition model to obtain a recognition result of the target image; the recognition result includes whether there is normal sperm and the position of the sperm; the target image is an image of any micro-pit; According to the recognition results of each image, the location and number of normal sperm on the microfluidic chip are determined; The target recognition model is improved based on the YOLOv8 model. The specific improvements include: Keep the first CBS module in the backbone structure and replace the remaining CBS modules with SPDConv modules; Before the SPPF module, a multi-scale feature enhancement module is inserted; An attention module is added between the jump connections of the trunk structure and the neck structure, combined with upsampling and channel concatenation to form a feature fusion module; Add a small object detection layer to the head structure.

2. The sperm identification method combining a microfluidic chip and machine vision according to claim 1, characterized in that: The multi-scale feature enhancement module includes multiple deep convolution branches; the calculation process of the multi-scale feature enhancement module includes: ； Among them, X0 is the input of the multi-scale feature enhancement module, X2, X3 and X4 are the outputs of the three deep convolution branches respectively; W is the attention weight; Y is the output of the multi-scale feature enhancement module; f is the operation symbol, the subscripts Conv and DWConv represent standard convolution and deep convolution respectively, and the superscript represents the convolution kernel size.

3. The sperm identification method combining a microfluidic chip and machine vision according to claim 1, characterized in that: The calculation process of the feature fusion module includes: Receive the feature map transmitted by the backbone structure and perform spatial attention weighting on it to obtain the first feature map; receiving a feature map transmitted by the neck structure and upsampling the feature map to obtain a second feature map; The first feature map and the second feature map are concatenated in the channel dimension to obtain and output the third feature map.

4. The sperm identification method combining a microfluidic chip and machine vision according to claim 1, characterized in that: During the model training process, the EIoU loss function is used as the bounding box regression loss of the detection head; the calculation formula of the EIoU loss function is: ； Among them, L EIoU is the desired EIoU loss; L IoU is the IoU loss, IoU is the intersection over union ratio of the predicted box and the real box; L dis is the distance loss, dc is the distance between the center of the predicted box and the center of the real box, W p and H p L is the width and height of the minimum bounding box of the predicted box and the real box respectively; asp is the edge length loss, dw and dh are the width difference and height difference between the predicted box and the real box respectively.

5. A sperm recognition system combining a microfluidic chip and machine vision, characterized in that: The microfluidic chip has a high-density micro-pit array; the system comprises: A sample acquisition module, used to place the sperm sample in the microfluidic chip; An image acquisition module is used to acquire the image of each micropit in the microfluidic chip after the sperm has settled, and encode it so that each image corresponds to a position of a horizontal and vertical coordinate; The sperm recognition module is used to use the target image as the input of the pre-trained target recognition model to obtain the recognition result of the target image; the recognition result includes whether there are normal sperm and the sperm position; the target image is an image of any micro-pit; the target recognition model is improved based on the YOLOv8 model, and the specific improvements include: Keep the first CBS module in the backbone structure and replace the remaining CBS modules with SPDConv modules; Before the SPPF module, a multi-scale feature enhancement module is inserted; An attention module is added between the jump connections of the trunk structure and the neck structure, combined with upsampling and channel concatenation to form a feature fusion module; Add a small target detection layer to the head structure The information integration module is used to determine the location and quantity of normal sperm on the microfluidic chip based on the recognition results of each image.

6. The sperm recognition system combining a microfluidic chip and machine vision according to claim 5, characterized in that: The multi-scale feature enhancement module includes multiple deep convolution branches; the calculation process of the multi-scale feature enhancement module includes: ； Among them, X0 is the input of the multi-scale feature enhancement module, X2, X3 and X4 are the outputs of the three deep convolution branches respectively; W is the attention weight; Y is the output of the multi-scale feature enhancement module; f is the operation symbol, the subscripts Conv and DWConv represent standard convolution and deep convolution respectively, and the superscript represents the convolution kernel size.

7. The sperm recognition system combining a microfluidic chip and machine vision according to claim 6, characterized in that: The calculation process of the feature fusion module includes: Receive the feature map transmitted by the backbone structure and perform spatial attention weighting on it to obtain the first feature map; receiving a feature map transmitted by the neck structure and upsampling the feature map to obtain a second feature map; The first feature map and the second feature map are concatenated in the channel dimension to obtain and output the third feature map.

8. The sperm identification system combining a microfluidic chip and machine vision according to claim 6, characterized in that: During the model training process, the EIoU loss function is used as the bounding box regression loss of the detection head; the calculation formula of the EIoU loss function is: ； Among them, L EIoU is the desired EIoU loss; L IoU is the IoU loss, IoU is the intersection over union ratio of the predicted box and the real box; L dis is the distance loss, dc is the distance between the center of the predicted box and the center of the real box, W p and H p L is the width and height of the minimum bounding box of the predicted box and the real box respectively; asp is the edge length loss, dw and dh are the width difference and height difference between the predicted box and the real box respectively.

Images