Underwater fish target detection method and device based on R2Net, and storage medium

Abstract

The invention provides an underwater fish target detection method and a device based on R2Net, and a storage medium, and relates to the technical field of target detection. When underwater fish target detection is carried out, the method comprises steps: firstly, acquiring an underwater fish image for training; inputting the underwater fish image for training into a complex network structure of an R2Net model for training to obtain network parameters of the complex network structure of the R2Net model; performing equivalent transformation on the complex network structure of the R2Net model according to the network parameters of the complex network structure of the R2Net model to obtain a simplified network structure of the R2Net model; wherein the simplified network structure of the R2Net model is a Kmax * Kmax equivalent convolution kernel; obtaining a to-be-detected underwater fish image; inputting the to-be-detected underwater fish image into the simplified network structure of the R2Net model to obtain a feature extraction result; and performing underwater fish target detection based on the feature extraction result to obtain an underwater fish target detection result. According to the invention, the R2Net model is used for underwater fish target detection, and high-precision rapid underwater fish target detection is realized.

Description

technical field [0001] The present invention relates to the technical field of target detection, in particular to the R-based 2 Net's underwater fish target detection method, device and storage medium. Background technique [0002] In the process of underwater fish target detection, the backbone network is the most important part of the target detection network. The backbone network is the network responsible for feature extraction in the target detection network. It is generally used for the front-end to extract image features of different levels and scales, and generate feature maps for subsequent networks. The emergence of Resnet has changed the defect that the previous backbone network (such as Alexnet, VGG, etc.) continued to increase the number of layers after reaching a certain depth, resulting in a decrease in performance. Underwater images have blurred image details, and there are few available feature details. Therefore, in the target detection network, a more po...

AI Technical Summary

Problems solved by technology

[0004] However, Res2Net replaces the 3x3 convolution module in the original ResNet structure with a complex structure that considers multi-scale characteristics. Although the detection accuracy of the model is effect

Images