Study On Family Identification And Source Tracing Of Eriocheir Sinensis Based On In-depth Learning

With the rapid development of deep learning technology in the field of object detection and biometrics,typical applications such as smart pig breeding,fruit tree injury detection,and wild animal trace detection have emerged in recent years.Chinese mitten crabs,also known as river crabs,hairy crabs,is a rarity in China’s aquaculture industry,with greater economic value.However,because the shell characteristics of different crab strains are very similar,it is difficult for non-professionals except aquatic experts to distinguish the crab strains according to the shell characteristics,which greatly affects the selection and breeding of river crabs.In addition,due to the differences in water quality,temperature,soil and other factors in various breeding sites,the quality of river crabs cultivated in different places is greatly different.Because of the excellent geographical location,some producing areas have formed famous brands,such as "Yangcheng Lake hairy crab",which has brought sales advantages.At the same time,it also leads to a lot of fake and shoddy cases.In order to protect the legitimate rights and interests of consumers and the brand’s own image,it is of great value to propose a method that can distinguish the strains of river crabs and accurately trace the origin of individual river crabs.In this study,according to the shell characteristics of Chinese mitten crab,deep learning and image processing related technologies were used to study the strain identification and traceability technology of Chinese mitten crab.The main research contents are as follows:Firstly,the data sets of Chinese mitten crab strain identification and traceability were created.The crab shell image data were collected based on the field fishing of river crabs,and these data were manually labeled using label Img software.Second,aiming at the problem that different strains of Chinese mitten crab have high similarity in shell characteristics,and it is difficult for farmers to directly distinguish crab strains,a Chinese mitten crab strain recognition algorithm named YOLOCrab based on YOLOv5 s was proposed.Firstly,by integrating the Swi-Transformer module in the backbone network of YOLOv5 s,the correlation degree between each pixel of the crab shell is improved.Secondly,in order to alleviate the problem that the self-attention mechanism does not pay Attention to the importance of the image region,a SA(Shuffle Attention)channel random mixed attention module is added in front of each detection head in the neck network of YOLOv5 s,which effectively improves the model’s attention to the detected object in the image.The detection rate of Chinese mitten crab in the image is improved.Finally,by reconstructing the loss function of YOLOv5 s,the target prediction box is optimized and the convergence speed of the model is accelerated.The experimental results show that the accuracy,recall and mean Average Precision(m AP)of the improved algorithm in the Chinese mitten crab strain identification reach 94.2%,94.8% and 98.5%respectively.The algorithm proposed in this study can quickly and accurately identify the strain of Chinese mitten crab,and provide an effective technical means for the sustainable and healthy development of freshwater aquaculture production of Chinese mitten crab.Third,the problem of individual identification and traceability of Chinese mitten crab.By randomly adjusting the contrast,brightness,gray level of the image,and affine changes,120 enhanced images are generated for each crab,which effectively expands the crab data set.Then based on Face Net network model,aiming at the problem that its backbone network has a large amount of calculation,which leads to too slow detection speed and cannot be deployed on devices with low computing power,a lightweight backbone network Conv Ne Xt-T is integrated into Face Net network model.The experimental results show that the improved network can still distinguish the crabs used in the test with an accuracy rate of 100% while greatly improving the detection speed,which provides a feasible scheme for the traceability of river crabs.