Night underwater fish target detection method

Abstract

The invention discloses a night underwater fish target detection method. The target detection method comprises the steps: acquiring night underwater fish images according to a certain time interval; performing image enhancement processing on the night underwater fish image by using an MSRCP algorithm to obtain an enhanced night underwater fish image; inputting the enhanced night underwater fish image into a trained DetNASNet backbone network to carry out feature information extraction; inputting the extracted feature information into a Cascade R-CNN model optimized by a Soft-NMS algorithm to carry out target detection; and obtaining the night underwater fish target detection result. According to the method, adverse effects of illumination in a night environment on target detection are avoided, the color quality of underwater fish images is improved, repeated candidate boxes can be effectively removed, and more efficient target detection backbone networks are utilized, so that target detection of underwater fishes at night is more accurate and efficient.

Description

technical field [0001] The invention relates to the field of video target detection, in particular to the field of underwater fish target detection at night. Background technique [0002] At present, the detection algorithm of moving objects is relatively mature. The traditional machine learning method is to extract fish features through manual marking, and then input the features into classifiers such as support vector machines for detection and classification. The current extraction methods are: deformable extraction technology based on template matching is used for fish detection; the ratio of each color component in the fish image to the long-short axis is used as the fish feature; the image of each part of the fish is segmented from the fish image module, and then extract features with higher correlation. However, the above extraction method is very cumbersome and requires manual design of corresponding rules for each feature of fish, which is not suitable for detectin...

AI Technical Summary

Problems solved by technology

However, the above methods have low target detection accuracy in objective environments such as uneven illumination at night, excessive noise, irregular fish movement, and low video quality.

[0004] Many target detection methods that are relatively mature on land are not well suited for underwater environments

In addition, the underwater environment at night is unevenly illuminated, and the captured images are blurry and the scene is dark, which makes the target detection task more difficult.

Images