| 摘要: |
| 为准确、快速、全面获取内陆养殖小区池塘及尾水处理池水体水质变化情况，建立养殖水环境实时预警调控及数字化管控机制，选择浙江湖州市集中连片的养殖小区为试验区，于2020年12月采用搭载GaiaSky-mini高光谱相机的无人机进行试验区近地遥感图像采集，并进行图像拼接、辐射校正和几何校正等预处理；然后对反射波段进行差值、比值及归一化差值指数进行数值变换，基于相关性分析筛选不同水质参数的敏感波段，进而采用线性函数、指数函数和多项式函数构建各水质参数定量反演模型，并用全波段构建偏最小二乘回归（PLS）、径向基神经网络（RBF）和支持向量机（SVM）反演模型，进行模型验证与评价；最后基于各水质最佳水质模型对试验区水质参数进行空间分布反演与分析。结果表明，总悬浮物（TSS）、总氮（TN）、总磷（TP）、高锰酸盐指数（CODMn）和氨氮（NH4+）的最大相关系数分别为0.86、0.65、0.72、–0.85和0.75；对比6种建模方法，多项式函数对TSS反演精度最高，偏最小二乘对NH4+反演精度最高，RBF对TN和CODMn反演精度最高，支持向量机对TP反演精度最高，其中NH4+反演精度需要进一步优化提高，其他4种水质参数均达到养殖池塘预测精度要求。基于无人机高光谱技术可快速反演养殖小区水质空间分布图，实现养殖池塘及尾水处理池水质信息的快速准确获取。 |
| 关键词: 无人机 高光谱 养殖池塘 机器算法 水质反演 |
| DOI：10.11693/hyhz20210600133 |
| 分类号:S951.2;S966;X87 |
| 基金项目:浙江省公益技术研究计划项目，LGN20C190004号；国家重点研发计划项目，2019YFD0900302号。 |
|
| WATER QUALITY MONITORING OF INLAND AQUACULTURE PONDS BASED ON UAV HYPERSPECTRAL REMOTE SENSING TECHNOLOGY |
|
LIU Mei1, MA Qi-Liang2, YUAN Ju-Lin1, NI Meng1, LIAN Qing-Ping1, GUO Ai-Huan1
|
|
1.Agriculture Ministry Key Laboratory of Healthy Freshwater Aquaculture, Key Laboratory of Fish Health and Nutrition of Zhejiang Province, Zhejiang Institute of Freshwater Fisheries, Huzhou 313001, China;2.Center of Information Technology, Huzhou University, Huzhou 313000, China
|
| Abstract: |
| To accurately, quickly, and comprehensively monitor the changes in water quality of inland aquaculture ponds and wastewater treatment ponds of a farming community, a real-time early warning and digital control system was established. In December 2020, a farming community in Huzhou City, Zhejiang Province in China was selected to conduct the experimental research. An unmanned aerial vehicle (UAV) equipped with a GaiaSky-mini hyperspectral camera was used to collect near-ground remote sensing images of the area, and these images were pre-processed in technologies including image stitching, radiation correction, and geometric correction. Next, the reflectance bands were transformed numerically using difference, ratio, and normalized difference index. The sensitive bands of different water quality parameters were determined based on correlation analysis results. Linear, exponential, and polynomial functions were used to construct quantitative inversion models for each water quality parameter. Partial least squares (PLS) regression, radial basis function (RBF) networks, and support vector machine (SVM) inversion models were constructed, verified, and evaluated in the full bands. Finally, the spatial distribution of water quality parameters in the study area was predicted and analyzed based on the best water quality model available. The results show that the maximum correlation coefficients of total suspended solids (TSS), total nitrogen (TN), total phosphorus (TP), potassium permanganate index (CODMn), and ammonium-nitrogen (NH4+) were 0.86, 0.65, 0.72, –0.85, and 0.75 respectively. Comparing the six modeling methods, the model constructed via polynomial functions had the highest accuracy for TSS inversion, the PLS model had the highest accuracy for NH4+ treatment inversion, the RBF model had the highest accuracy for TN and CODMn inversion, and the SVM model had the highest accuracy for TP inversion. However, the accuracy of NH4+ inversion requires further optimization, while the other four water quality parameters all met the prediction accuracy requirements. Based on these results, a spatial distribution map of water environment in the farming area can be established via UAV hyperspectral technology, with which a rapid and accurate evaluation of water quality in farming and tailwater treatment ponds could be achieved. |
| Key words: UAV hyperspectral technology inland ponds machine algorithm water quality inversion |
