Abstract:

Black and odorous water bodies with small areas, hidden locations, or sudden occurrences are difficult to effectively identify using the band ratio method and deep learning technology. To solve this problem, this study proposes a comprehensive remote sensing identification method driven by high-scoring remote sensing images that identifies and judges black and odorous water bodies based on spatial distribution patterns, pollution sources, and remote sensing characteristics, as well as remote sensing identification methods for domestic sewage, domestic garbage, aquaculture manure, and industrial sewage. In this study, 16 districts and counties in Cangzhou City were used as research areas for remote-sensing identification and verification of black and odorous water bodies. The results showed that 1) 137 black and smelly water bodies were identified in the study area from three aspects: distribution location, formation reason, and image characteristics. 2) After field verification, the precision, recall, and accuracy rates of black and odorous water bodies were 65.8%, 97.1%, and 64.6%, respectively. The F1 Score was 0.78, the wrong classification rate was 34.2%, and the missing classification rate was 1.9%, which shows that this method can effectively identify black and odorous water bodies in cities. 3) Through the analysis of the operation time, it was found that this method can obtain the spatial distribution position of black and odorous water over a large area in a short time. This technology not only provides rapid and stable black and odorous water investigation services for government departments but also provides technical support for urban ecological environment improvement

Key words: black and odorous water body, knowledge-driven, high-scoring remote sensing, classification and recognition

摘要：

面积较小、位置隐蔽或突发性的黑臭水体难以通过波段比值法和深度学习技术进行有效地识别。针对这一问题，提出了一种基于高分遥感影像的、以知识为驱动的综合遥感识别方法。该方法从空间分布模式、污染源和遥感特征3个方面对黑臭水体逐步进行识别与黑臭判定，按照生活污水类、生活垃圾类、养殖粪污类和工业污水类4个类别开展分类综合研究，给出了每一类适用的遥感识别方法。以沧州市为研究区，开展疑似黑臭水体遥感识别。研究结果表明：1）基于知识驱动的疑似黑臭水体识别方法有效，通过分布位置、形成原因和影像特征3个方面在研究区识别出137处黑臭水体；2）经过精度验证，黑臭水体查准率为65.8%，查全率为97.1%，准确率为64.6%，F1 Score为0.78，错分率为34.2%，漏分率为1.9%，表明该方法能够有效全面地识别黑臭水体；3）作业耗时分析，发现该方法能够在较短的时间内获得大范围地区的黑臭水体分布位置。因此，该技术可为政府部门提供迅速的稳定的黑臭水体排查服务，为生态环境改善提供技术支持。

关键词: 黑臭水体, 知识驱动, 高分遥感, 分类识别