关键词: 细粒度分类, 深度学习, 特征融合

Abstract: In the national land use change survey,the manual visual interpretation of field-photographed verification photos is inefficient.Identifying crop types to determine land use types is a key approach to solve this problem.However,crop type identification faces challenges such as large intra-class differences and small inter-class differences.This paper proposes a fine-grained crop classification algorithm based on the fusion of global and local features.Global features are extracted from the original images,and local features are obtained by using a detector to acquire target regions.After feature compression and fusion,the features are input into a classifier.The exponential moving average method is used to update network parameters,and the additive angular cross entropy loss function is selected to complete model training.Experiments on real-world data from the national land use change survey show that this algorithm outperforms comparative algorithms in terms of precision,effectively improving the accuracy of fine-grained crop classification and providing efficient intelligent interpretation technical support for related businesses such as the national land use change survey.

Key words: fine-grained crop classification, deep learning, feature fusion