摘要
蟹塘中的水草不仅能够净化水质、增加溶氧量、提供饵料,夏季高温期还可为螃蟹提供遮阳和栖息的场所。目前,水产养殖过程中水草种植信息的获取主要依靠人工判断,存在劳动力成本高、时效性差以及效率低等问题。为了有效解决以上问题,本研究选取南京市高淳区和无锡市宜兴市大闸蟹养殖塘作为研究区,将卷积神经网络与无人机高空间分辨率多光谱影像相结合,构建基于无人机多光谱影像的多时相蟹塘水草分布数据集,提出基于残差多尺度提取网络的水草信息精细提取模型(ResMSFCN),并与U-Net模型和DeepLabv3+模型进行精度比较。结果表明,ResMSFCN模型具有较好的水草影像提取能力,准确率达0.963,交并比达0.804。此外,该模型能精准识别6月份至10月份5个时相的水草分布区域均匀度。本研究提出的基于无人机多光谱影像和ResMSFCN结合的蟹塘水草信息提取模型能高精度、高效率获取蟹塘多时相水草覆盖信息,为蟹塘水草资源快速评估提供科学依据。
Aquatic plants in crab ponds not only purify water quality,increase dissolved oxygen,and provide food,but also offer shading and habitats for crabs during the high-temperature period in summer.Currently,the acquisition of aquatic plant planting information in aquaculture mainly relies on manual judgment,which has problems such as high labor costs,poor timeliness,and low production efficiency.To effectively solve these issues,this study selected crab ponds for Chinese mitten crabs in Gaochun District of Nanjing and Yixing City of Wuxi as the research areas.Combining convolutional neural networks with high-spatial-resolution multispectral images acquired by unmanned aerial vehicles(UAVs),a multi-temporal dataset of aquatic plant distribution in crab ponds based on UAV multispectral images was constructed.A fine extraction model for aquatic plants based on the residual multiscale fully connected neural network(ResMSFCN)was proposed,and its accuracy was compared with U-Net and DeepLabv3+models.The results showed that the ResMSFCN model had better ability to extract aquatic plant images,with an accuracy of 0.963 and an intersection over union(IoU)of 0.804.In addition,this model could accurately identify the uniformity of aquatic plant distribution areas across five temporal phases from June to October.The model proposed in this study,which combines UAV multispectral images with ResMSFCN for aquatic plant extraction in crab ponds,can acquire multi-temporal aquatic plant coverage information with high precision and efficiency,and also offer a scientific basis for the rapid evaluation of aquatic plant resources in crab ponds.
作者
董建宾
李卫国
戴佩玉
毛星
金晶
DONG Jianbin;LI Weiguo;DAI Peiyu;MAO Xing;JIN Jing(School of Ecology and Applied Meteorology,Nanjing University of Information Science&Technology,Nanjing 210044,China;Institute of Agricultural Information,Jiangsu Academy of Agricultural Sciences,Nanjing 210014,China)
出处
《江苏农业学报》
北大核心
2026年第2期349-356,共8页
Jiangsu Journal of Agricultural Sciences
基金
江苏省重点研发计划项目(BE2022366)
江苏现代农业产业单项技术研发项目[CX(23)3131]
关键词
蟹塘
水草
卷积神经网络
无人机遥感
多光谱影像
crab pond
aquatic plants
convolutional neural network
unmanned aerial vehicle remote sensing
multispectral images
