Research On Freshwater Fish Species Recognition And Key Point Location Technology Based On Deep Learning

Freshwater fish has tender meat and rich nutrition,which is one of the important protein sources for people in daily life.At present,aquatic products are basically supplied to the market in fresh and live form.Due to the lag in freshwater fish processing technology and the limited capacity of the fresh market,the income of the freshwater fish farming industry has decreased and the development of the freshwater fish industry has been restricted.In order to further improve the economic value of freshwater fish products,it is the only way to study the deep processing technology of freshwater fish.Before deep processing,sorting and pre-processing freshwater fish is an important step.In the pre-treatment and processing of freshwater fish,manual identification,selection and gill removal are mainly relied on,which is labor intensive and low in accuracy.With the development of computer technology,the application of deep learning in many fields has made major breakthroughs,especially in the field of automatic classification and identification of agricultural products.This paper uses the advantages of deep learning to learn from itself and resist external interference,and carry out research on freshwater fish species identification and key point location technology based on deep learning.The main research results obtained in this paper are as follows:(1)Aiming at the poor working environment and low recognition accuracy of traditional manual fish sorting,this paper proposes an image recognition method combining SPP(Spatial Pyramid Pooling)and DenseNet,by adding SPP in front of the fully connected layer of the DenseNet network,the over-fitting of the data is reduced and the classification accuracy is improved.The test results of 6 kinds of freshwater fish images show that the classification accuracy of this method is higher than that of unmodified DenseNet,VGG-16,ResNet50 and Inception V3,and it can effectively identify freshwater fish images.(2)In order to improve the accuracy of gill cut detection and positioning,this paper proposes a method for detecting and positioning freshwater fish gill cuts based on the improved Faster RCNN.First,a batch normalization layer is added to the backbone network VGG16 to enhance the network feature extraction capability;then,the Soft-NMS algorithm is used to replace the NMS(Non-Maximum Suppression)algorithm in the candidate area network to improve the accuracy of small target detection.Through comparative experiments with other fish gill cut detection algorithms,the results show that the F1 value and positioning effect of the improved Faster RCNN network are higher than other algorithms,laying the foundation for the subsequent production process.(3)In binocular vision ranging,firstly,the internal and external parameters of the camera are obtained to correct the image;then,the corrected image is registered using a stereo matching algorithm to obtain the spatial coordinate value of the center point of the gill cut detection frame.The three-dimensional positioning of the gill incision point is realized.(4)On the basis of the above research results,in the PyCharm development environment,using PyQt5 to design freshwater fish species identification and key point location software.The upload image test results show that the software achieves accurate classification of freshwater fish species and precise positioning of fish gill cut points.