基于U-Net神经网络的浮筏养殖信息提取——以长海县为例  

Extraction of Floating Raft Aquaculture Information Based on U-Net Neural Network——A Case Study of Changhai County

作　　者：由金浩 刘威[1] 王权明 You Jihao;Liu Wei;Wang Quanming(School of Geographical,Liaoning Normal University,Dalian 116029,Liaoning China;National Marine Environmental Monitoring Center,Dalian University of Technology,Dalian,116023,Liaoning,China)

机构地区：[1]辽宁师范大学地理科学学院,辽宁大连116029 [2]国家海洋环境检测中心,辽宁大连116023

出　　处：《绿色科技》2024年第2期261-265,共5页Journal of Green Science and Technology

摘　　要：浮筏养殖是海水养殖中最重要的类型之一,对于浮筏养殖的精确提取尤为重要。然而浮筏在遥感影像中分布密集,且背景复杂,并且以往的浮筏提取方法,多为机器学习,未能挖掘浮筏养殖的深度特征,以及光谱信息的高效利用。针对上述问题,提出了U-Net网络模型进行浮筏养殖信息提取,使用比值指数计算特征波段,去除冗余光谱信息,并添加深度神经网络,深化提取浮筏信息提取的深度特征通道,实现了浮筏养殖信息的高精度提取。选取长海县大长山岛作为研究区域,与Canny算子、Otsu算法、PCA_Kmeans算法提取结果进行了对比,U-Net模型浮筏养殖的提取总体精度为95.6%,与常用的机器学习算法相比,提取精度提高了9%~13%,验证了U-Net模型在浮筏养殖识别中的高效性。Floating raft aquaculture is one of the most important types of marine aquaculture,so accurate extraction of floating raft cultivation is particularly important.However,floating rafts are densely distributed in remote sensing images and have complex backgrounds.Previous methods for extracting floating rafts mainly rely on machine learning,which fail to explore the deep features of floating raft cultivation and efficiently utilize spectral information.To address these issues,a U-Net network model for extracting floating raft aquaculture information is proposed.Ratio indices are used to calculate feature bands,remove redundant spectral information,and add a deep neural network to deepen the extraction of deep feature channels for floating raft information,achieving high-precision extraction of floating raft aquaculture information.Dachangshan Island in Changhai County is selected as the study area and the extraction results with Canny operator,Otsu algorithm,and PCA_Kmeans algorithm is compared.The overall accuracy of floating raft aquaculture extraction using the U-Net model is 95.6%,which is 9% to 13% higher than common machine learning algorithms,demonstrating the efficiency of the U-Net model in identifying floating raft aquaculture.

关 键 词：浮筏养殖 U-Net模型 比值指数 深度特征 光谱信息 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]