Fine-grained bird image retrieval method based on graph neural network and deep hash

Abstract

The invention discloses a fine-grained bird image retrieval method based on a graph neural network and deep Hash, which belongs to the field of fine-grained image retrieval, and comprises the following steps: node representation based on local features, local feature enhancement, related part relation mining based on graph convolution, semantic Hash coding and loss function. The invention provides a comprehensive method suitable for large-scale bird image retrieval based on a graph neural network and a deep hash method, and fine-grained bird image retrieval can be realized in a high-efficiency, low-storage and high-precision manner.

Description

technical field [0001] The invention belongs to the field of fine-grained image retrieval, in particular to a fine-grained bird image retrieval method based on a graph neural network and deep hashing. Background technique [0002] At present, for nature reserves and national wetland areas, bird monitoring can be used as an important indicator of biodiversity and ecological environment evaluation; for farmland and airport areas, bird monitoring affects the economic income of farmers and the normal operation of the airport. In nature, there are many kinds of birds. Therefore, how to quickly and accurately retrieve the correct bird from a large-scale data set is what bird experts hope. Because the retrieval of bird images is different from general image retrieval, there are two characteristics between its database images and query images: (1) Small differences between classes: the differences between birds are small, and the differences are often subtle. For example, a bird he...

AI Technical Summary

Problems solved by technology

(2) There is a large difference within the class. Due to factors such as illumination and posture, different images of the same bird have great differences

However, the encoding dimension of these two methods is 1024 dimensions, and its encoding dimension is relatively high. In the actual large-scale image retrieval, it may encounter the problems of slow query speed and redundant storage.

Images