海洋渔业2025，Vol.47Issue(6)：842-853,12.
基于SPD-BsiNet的高分遥感影像养殖池塘提取方法
Aquaculture pond extraction method based on SPD-BsiNet from high-resolution remote sensing images
摘要
Abstract
Aiming at solving the problems of insufficient segmentation accuracy and generalization performance of high-resolution remote sensing images in aquaculture pond extraction,a multi-task aquaculture pond extraction method SPD-BsiNet based on multi-scale enhancement was proposed.This method performed deep supervision on each stage of the decoder to achieve simultaneous learning of the intermediate features of different sub-tasks.Based on the dilated convolution,a dilated convolution unit and the pyramid feature extraction unit were introduced to increase the effective receptive field and improve the positioning ability of the model for different scale targets in complex interfering scenarios.The experimental results on the data sets of Jiuzhen Bay and Zhao'an Bay showed that the IoU,F1-score and CES indices of this method reached 0.889 6,0.941 6 and 0.047 9,respectively,which were better than those of D-Linknet,Densenet-Unet,Psi-Net and BsiNet.It could take into account the extraction effects of different types of aquaculture ponds and had better attribute accuracy and geometric accuracy.At the same time,based on the optimal weight obtained by training,this method had also achieved good application results in the extraction of unsampled areas such as Zhao'an Bay,Fotan Bay and Jiulong River Estuary,and had strong generalization performance and practical value.关键词
养殖池塘/深度监督/膨胀卷积/SPD-BsiNetKey words
aquaculture ponds/deep supervision/expanded convolution/SPD-BsiNet分类
天文与地球科学引用本文复制引用
彭俊,陈红梅,陈芸芝,罗冬莲..基于SPD-BsiNet的高分遥感影像养殖池塘提取方法[J].海洋渔业,2025,47(6):842-853,12.基金项目
福建省海洋服务与渔业高质量发展专项资金项目(FJHY-YYKJ-2024-1-14) （FJHY-YYKJ-2024-1-14）
福建省省属公益类科研院所基本科研专项(2023R1012005) （2023R1012005）
福建省自然科学基金项目(2022J01111) （2022J01111）
