Research And Application Of River Crab Individual Recognition Algorithm Based On Image Features

In recent years,due to the rapid growth of my country’s national economy and the rise of the e-commerce industry,coupled with the vigorous development of logistics and express services,Chinese mitten crab,as one of my country’s characteristic aquatic economic crops,already has a large-scale sales market.However,the current mainstream hairy crab tracing method in the market is to use electronic labels such as bundling barcodes or two-dimensional codes on the crab claws.Due to the replaceability of the markers,it is not reliable to trace crab information by this method alone.Due to changes in the growth environment of hairy crabs,there will be obvious differences in the morphological traits such as uplift,depression,ditches,and textures in the carapace images of individual hairy crabs.Therefore,the morphology of individual crabs tends to be different.In view of the current high cost of river crab traceability and the inability of consumers to trace individual river crab information in a fine-grained manner,this paper proposes two individual river crab traceability algorithms based on the image of the carapace of river crabs.Finally,the feasibility and practicability of the crab image traceability algorithm are actually verified through the traceability application system.To this end,the main work of this paper is as follows:First,the task of this research is to obtain the unique morphological eigenvalues of individual crabs from the carapace images of river crabs.The crab image data is taken from a total of 100 adult hairy crabs in Yangcheng Lake.In order to better adapt the data to the real lighting conditions,the data collection site is selected in a bright room(fluorescent light or natural light).Since the shooting time includes morning,noon,evening and night,the collected images need to adapt to various reproduction lighting conditions.During the collection process,the MER-2000-5GC/P industrial camera was used to collect data,and the images needed to show clear and complete outlines,ravines,bumps and textures of the crab carapace.The first method proposed in this paper is an individual crab identification algorithm based on the fusion of HOG and LBP features.Since the HOG feature extraction algorithm alone can only extract the relevant directional information of the edge contour of the crab image,for the detailed information such as the gully texture on the carapace of the crab There is a lack of sufficient description,and the texture features extracted by the LBP algorithm descriptor alone are limited,and the ability to express the edge information of the image is poor.Therefore,a method of combining the two feature extraction algorithms for feature extraction is proposed,and the principal component analysis method is applied.The dimensionality reduction of the extracted features after fusion is performed,and finally,the support vector machine is used to classify and identify the single crab image.The results show that the feature fusion algorithm using the principal component analysis method can greatly reduce the feature dimension,and the recognition accuracy is also higher than other comparison algorithms;and the feature extraction algorithm using feature fusion is also higher than the recognition rate using LBP alone.Or the algorithm of HOG feature extraction.Therefore,while reducing the feature dimension,this algorithm improves the accuracy of single crab identification,which is of great significance for the use of crab images for information traceability.The second method proposed in this paper is an individual recognition algorithm of crab carapace images based on transfer learning and pyramid convolution.The algorithm uses pyramid convolution layers to replace ordinary residual convolution blocks to construct a network model,which can extract multi-scale and deep feature information from crab back images.The experimental results show that the accuracy rates of Resnet34 and Resnet50 with pyramid convolution structure are 98.38% and98.51%,respectively.Compared with the model using ordinary convolution layer,the accuracy rate is improved by 5.49% and 1.3%,and when the model depth reaches 101%The model performance is no longer significantly improved when layers are added.Compared with the new learning model using the pyramid convolution structure,the number of training convergence iterations of the transfer learning method is reduced from 20 to 5.At this time,the accuracy of the model is 98.88%,which is 0.37% higher than that of the new learning.make up for the small sample size.This study provides theoretical basis and technical support for individual identification and tracing of river crabs.Finally,in order to verify the effectiveness of the two proposed river crab individual identification algorithms,this paper designs and develops an individual river crab image identification and traceability system according to actual needs.The results show that the river crab traceability system based on river crab images is practical and feasible.The information digital management of the supply chain of crab products has a driving effect.