基于卷积神经网络的水产养殖区提取  

Extraction of Aquaculture Area Based onConvolutional Neural Network

作　　者：刘滨洪 LIU Binhong(6197 Geological Brigade of Fujian Province,Quanzhou 362011,China)

机构地区：[1]福建省197地质大队,福建泉州362011

出　　处：《测绘与空间地理信息》2023年第11期74-77,共4页Geomatics & Spatial Information Technology

摘　　要：传统的水产养殖信息统计是通过管理部门实地观察获取,时间效率和信息可靠性都较低,难以满足现实研究需要,而采用遥感技术结合卷积神经网络分类方法进行水产养殖区提取的方法能解决这一难点。本文根据哨兵二号影像上目标地物的特征修改SegNet卷积神经模型的结构,提出一种简化的卷积神经网络模型S-SegNet,实现对近海养殖区遥感高精度自动识别。结果表明,S-SegNet模型的分类效果相比SegNet模型有了明显的提高,平均准确度、平均召回率和平均F1-score分别达到88%、0.9和0.91,为三沙湾水产养殖区的实时监测和科学规划提供了重要数据支撑。While the traditional aquaculture information statistics obtained through the field observation of the management department is difficult to meet the needs of practical research because of the low time efficiency and information reliability,the method of remote sensing technology combined with convolution neural network classification for aquaculture area extraction can address the problem.In this paper,the structure of SegNet convolution neural network model is modified according to the characteristics of target objects on Sentinel-2 images,and a simplified convolution neural network model S-SegNet is proposed to realize high-accuracy automatic recognition of offshore aquaculture area.The results show that the classification effect of S-SegNet model is significantly improved compared with SegNet model,and the average accuracy,average recall and average F1-score are 88%,0.9 and 0.91 respectively,which provides important data support for real-time monitoring and scientific planning of Sansha Bay aquaculture area.

关 键 词：三沙湾 水产养殖 卷积神经网络 哨兵二号 SegNet 

分 类 号：P237[天文地球—摄影测量与遥感]