Lightweight underwater target detection method, system, medium, equipment and terminal

Abstract

The invention belongs to the technical field of underwater target detection, and discloses a lightweight underwater target detection method and system, a medium, equipment and a terminal. A CSPDarknet18 is used as a backbone network to preliminarily extract features; extracting image features of different levels and scales by using a Dense strategy; an AFF module is added between the backbone feature extraction network and the FPN structure to realize cross-channel information interaction; constructing FPN and PANet networks, and extracting features with semantic information and position information at the same time; the enhanced feature extraction network is reconstructed by using depth-divisible convolution instead of common convolution, so that the number of parameters is reduced. The method is mainly used for positioning and identifying images of sea cucumbers, scallops, sea urchins and starfishes. Experimental results show that the mAP of the method reaches 78.18% on a 2020 URPC underwater target detection data set, the model parameter size is 37.22 M, the processing speeds of the video data collected on site in the sea area of the islands are 10.95 FPS and 28.05 FPS respectively, and good effects are achieved in the aspects of accuracy and speed.

Description

technical field [0001] The invention belongs to the field of target detection, and in particular relates to a lightweight underwater target detection method, system, medium, equipment and terminal. Background technique [0002] With the ever-increasing demands on marine biomass production worldwide, bottom fishing technology is becoming more and more important. At present, the most commonly used method of fishing for seafood is artificial diving. However, this traditional fishing method has a high risk factor, short operation time and great harm to the body. In response to these problems, research on the target detection technology of underwater marine life can allow robots to replace humans to complete the task of marine life fishing, and provide an effective way for the development and utilization of marine resources. However, the complexity of underwater environment and lighting conditions leads to blurred details and color distortion of underwater images; underwater bio...

AI Technical Summary

Problems solved by technology

[0007] (1) Due to the propagation of light in water, the clarity of underwater images will be reduced. Underwater images often have problems such as low light and dark contrast, chromatic aberration, insufficient light, blur and lack of details, and noise pollution, which affect the accuracy of underwater image target detection.

[0008] (2) Underwater targets are usually small and clustered, making the detection accuracy of underwater small targets and multi-scale targets poor

[0009] (3) Due to the limited storage and computing capacity of underwater system devices, large-scale detection networks cannot play a role in underwater environments, so existing biological target detection methods in marine environments are difficult to achieve the purpose of real-time detection

Images