Filter cutting method for convolutional neural network and automatic shellfish classification system

Abstract

The invention discloses a filter cutting method for a convolutional neural network, and the method comprises the steps: carrying out calculation and sorting of the importance of filters, cutting off the filters with lower importance, calculating the orthogonality measurement between the filters in a layer, selecting the related filters with relatively small orthogonality; and cutting off the filters with lower importance ranks, and re-initializing the cut filters. Therefore, the filter cutting method inhibits correlation between features, more attention is paid to orthogonal features, different directions in an activation space are captured, and the generalization ability of a classification model is improved. The invention further discloses an automatic shellfish classification system, and the accuracy of automatic classification of high-similarity shellfishes is improved especially for the problem that the high-similarity shellfishes are difficult to recognize.

Description

technical field [0001] The invention relates to the field of machine learning, in particular to a filter clipping method for a convolutional neural network and an automatic shellfish classification system. Background technique [0002] Classification in taxonomy follows taxonomic principles and methods, and classifies various groups of organisms into kingdoms, phylums, classes, orders, families, genus, and species. In practical applications, the image features of shellfish belonging to the same family have high similarity and the samples are unbalanced, which puts forward higher requirements for the study of shellfish classification. At present, the convolutional neural network (CNN) is widely used in the recognition of object types, but when it is directly applied to the classification of shellfish of the same family, due to the similar characteristics of shellfish of the same family, the unbalanced distribution of samples of different shellfish and the difficulty of sample...

AI Technical Summary

Problems solved by technology

At present, the convolutional neural network (CNN) is widely used in the recognition of object types, but when it is directly applied to the classification of shellfish of the same family, due to the similar characteristics of shellfish of the same family, the unbalanced distribution of samples of different shellfish and the difficulty of sample classification Imbalance problem, CNN recognition accuracy is low, and the recognition effect is poor

Images