Research On Sex Recognition Method Of Portunus Sinensis Based On Point Cloud Image Classification

Portunus trituberculatus in Zhoushan sea area is an important economic crab in China’s coastal area.However,due to the different nutritional components of male and female crabs,it is necessary to grade the sex of Portunus before marketing.The efficiency of traditional human eye recognition and manual classification is very low.Therefore,based on the current situation of intelligent fishery,this paper studies the method of sex recognition of Portunus sinensis based on point cloud image classification,which provides a theoretical basis for the environmental awareness of modern fishery classification equipment.The main work of this paper is as follows:(1)A dataset for gender classification of Portunus sinensis(PGCD)was built: in the public database,there is no data set for sex classification of Portunus sinensis,so relevant dataset need to be built.Firstly,this paper collects some original data sets from aquatic products processing companies;Secondly,the original data set is preprocessed by using the method of proportionally resizing and filling;Then the data set is randomly divided into the original training set and the original test set;Finally,in order to solve the problem of over-fitting that a small number of samples may bring to the network,or the problem of focusing on the majority of classes in model prediction caused by sample imbalance,this paper uses five data enhancement methods to expand the original data set,thus establishing a PGCD to provide a database for subsequent experiments.(2)A sex recognition algorithm of Portunus sinensis based on multiple groups of convolutional neural networks is proposed.Res Net is introduced into the algorithm to extract features from image blocks to reduce information loss in the process of feature extraction and make feature extraction more powerful;Then,the attention mechanism is used to replace the traditional pooling layer,so as to find the useful information of the input data more attentively;Finally,a series of parameters are adjusted to make the proposed MGCNN have the best classification performance.The experimental results show that the proposed method has good classification performance on PGCD.Although the method in this chapter is only a preliminary attempt to classify the image blocks of Portunus sinensis,it has well explained the effectiveness of multiple sets of convolution neural networks,and provided important theoretical support for the subsequent research on high-precision classification of Portunus sinensis point cloud images.(3)A sex recognition algorithm for Portunus sinensis based on multi-view learning aggregation network is proposed.Res2 Net is introduced into the algorithm to extract features from 2D views with multiple rotating views,which further increases the number of acceptable domains and makes the feature extraction ability more powerful;At the same time,attention pooling is used instead of the traditional maximum pooling,which focuses more on finding the useful information related to the current output in the input data,and effectively solves the problem of feature information loss caused by feature representation,as well as the loss of detail information in the dimensionality reduction process of each view.Experimental results show that compared with advanced methods,the network framework can achieve higher classification accuracy and better performance.Its advantages verify the effectiveness of multi-view learning aggregation network.