广东海洋大学学报2025，Vol.45Issue(5)：9-18,10.DOI:10.3969/j.issn.1673-9159.2025.05.002
基于图像增强的夜间低照度金枪鱼渔获物图像分类性能提升
Improvement of Classification Performance of Nighttime Low-Light Tuna Catch Images Based on Image Enhancement
摘要
Abstract
[Objective]This paper aims to research on an optimization method based on image enhancement to provide a solution for improving the classification accuracy of tuna longline catch images under low-illumination nighttime conditions.[Methods]A nighttime image optimization approach combining dynamic white balance correction and multi-scale Retinex enhancement was proposed.Three representative deep learning classification models,such as ResNet50,VGG16 and MobileNetV3,were employed to comprehensively evaluate the performance variations before and after enhancement in terms of classification accuracy and F1-score.[Results]This method can effectively suppress the green color cast in low light images,and improve the brightness balance and detail clarity of the images.Compared to raw nighttime images,the enhanced nighttime images achieve an average accuracy improvement of approximately 6%-8%across all three models.Specifically,ResNet50 accuracy increases from 0.86 to 0.92;VGG16 accuracy rises from 0.84 to 0.92;and MobileNetV3 from 0.82 to 0.87.The classification performance of enhanced nighttime images approached that under well-lit daytime conditions.[Conclusion]The effectiveness of image enhancement methods was validated in improving the classification performance of nighttime fishing catch images,and the result can expand the technical pathways for applying intelligent fishery monitoring technologies in complex lighting scenarios.关键词
金枪鱼延绳钓/图像增强/图像分类/深度学习/人工智能渔业Key words
tuna longline fishing/image enhancement/image classification/deep learning/artificial intelligence fisheries分类
农业科技引用本文复制引用
卢世航,黄小双,孔祥洪,陈新军..基于图像增强的夜间低照度金枪鱼渔获物图像分类性能提升[J].广东海洋大学学报,2025,45(5):9-18,10.基金项目
国家重点研发计划(2023YFD2401305) （2023YFD2401305）
国家自然科学基金(41876141) （41876141）
