摘要
为准确检测水下生物,本文采用优化的MSRCR算法对水下图像进行增强,并基于DenseNet思想提出一种改进的YOLOv3目标检测算法Den-YOLOv3。针对水下生物图像中存在的图像模糊、色偏严重的问题,将ACE算法作用于MSRCR算法增强后的图像,以优化图像颜色及亮度。同时,将YOLOv3特征提取网络中的残差模块替换成密集块,得到DenYOLOv3模型,该模型有效加强了特征传播,提高了检测精度。实验结果表明:优化后的ACE_-MSRCR算法能有效增强水下图像,进而提高检测精度。Den-YOLOv3对四类水下生物的检测精度均高于其他模型,在检测速度上达到25 ms/帧,可以满足对水下生物实时检测的要求。
In order to accurately detect underwater organisms,this paper adopts the optimized MSRCR algorithm to enhance underwater images,and proposes an improved YOLOv3 target detection algorithm Den-YOLOv3 based on the DenseNet idea. The ACE algorithm is applied to the enhanced MSRCR image to optimize the color and brightness of the image,aiming at the serious problems of image blurriness and color deviation. Meanwhile,the residual module in the YOLOv3 feature extraction network was replaced with dense blocks to obtain Den-YOLOv3 model,which effectively enhanced feature propagation and improved detection accuracy. The experimental results show that the optimized ACE_MSRCR algorithm can effectively enhance the underwater image and improve the detection accuracy. Den-YOLOv3 has a higher detection accuracy than other models for the four types of underwater creatures,and the detection speed reaches 25 ms/frame,which can meet the requirements for real-time detection of underwater creatures.
作者
郝琨
王阔
赵璐
王贝贝
王传启
HAO Kun;WANG Kuo;ZHAO Lu;WANG Bei-bei;WANG Chuan-qi(School of Computer and Information Engineering,Tianjin Chengjian University,Tianjin 300384,China;School of Control and Mechanical Engineering,Tianjin Chengjian University,Tianjin 300384,China;Tianjin Keyvia Electric Co.,Ltd.,Tianjin 30038Ay China)
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2022年第5期1088-1097,共10页
Journal of Jilin University:Engineering and Technology Edition
基金
国家自然科学基金项目(61902273)。
关键词
计算机应用
水下生物检测
水下图像
图像增强
深度学习
computer application
underwater biological detection
underwater image
image enhancement
deep learning
