多尺度特征融合下的高帧频图像关键目标识别  

Key Target Recognition in High Frame Rate Images Based on Multi-Scale Feature Fusion

作　　者：林琳[1,2] Lin Lin(Institute of Marine Mechanical and Electrical Engineering,Xiamen Ocean Vocational College,Xiamen,Fujian 361100,China;Xiamen Key Laboratory of Intelligent Fishery,Xiamen,Fujian 361100,China)

机构地区：[1]厦门海洋职业技术学院海洋机电学院,福建厦门361100 [2]厦门市智慧渔业重点实验室,福建厦门361100

出　　处：《黑龙江工业学院学报（综合版）》2024年第3期73-79,共7页Journal of Heilongjiang University of Technology(Comprehensive Edition)

基　　金：福建省海洋与渔业局海洋经济发展专项资金项目(项目编号:FJHJF-L-2021-12);福建省教育厅“智慧渔业应用技术协同创新中心”(项目编号:XTZX-ZHYY-1914)。

摘　　要：为优化高帧频图像目标识别效果,提出基于多尺度特征融合的高帧频图像关键目标识别方法。结合空域双边滤波算法和双树复小波变换算法去除高帧频图像噪声。通过多个卷积模块提取图像特征,并在分支空间注意力机制、改进通道注意力网络下融合多尺度特征。引入联合稀疏概念表征融合后的多尺度特征,并将其输入卷积神经网络结构中进一步学习,输出关键目标识别结果。实验结果表明,所提方法应用后AUC值为0.91,满足了高帧频图像处理要求。To optimize the target recognition performance of high frame rate images,a key target recognition method based on multi-scale feature fusion is proposed for high frame rate images.Combining spatial bilateral filtering algorithm and dual tree complex wavelet transform algorithm to remove noises from high frame rate images.Extract image features through multiple convolutional modules and fus multi-scale features under branch spatial attention mechanisms and improved channel attention networks.Introduce the concept of joint sparsity to represent the fused multi-scale features,and input them into the convolutional neural network structure for further learning,outputting key target recognition results.The experimental results show that the AUC value of the proposed method after application is 0.91,which meets the requirements of high frame rate image processing.

关 键 词：帧频图像 特征融合 空间注意力 空域双边滤波 目标识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]