Rapid Prediction and Inversion of Pond Aquaculture Water Quality Based on Hyperspectral Imaging by Unmanned Aerial Vehicles

Water quality in aquaculture has a direct impact on the growth and development of the aquatic organisms being cultivated. The rapid, accurate and comprehensive control of water quality in aquaculture ponds is crucial for the management of aquaculture water environments. Traditional water quality monitoring methods often use manual sampling, which is not only time-consuming but also reflects only small areas of water bodies. In this study, unmanned aerial vehicles (UAV) equipped with high-spectral cameras were used to take remote sensing images of experimental aquaculture ponds. Concurrently, we manually collected water samples to analyze critical water quality parameters, including total nitrogen (TN), ammonia nitrogen (NH4+-N), total phosphorus (TP), and chemical oxygen demand (COD). Regression models were developed to assess the accuracy of predicting these parameters based on five preprocessing techniques for hyperspectral image data (L2 norm, Savitzky–Golay, first derivative, wavelet transform, and standard normal variate), two spectral feature selection methods were utilized (successive projections algorithm and competitive adaptive reweighted sampling), and three machine learning algorithms (extreme learning machine, support vector regression, and eXtreme gradient boosting). Additionally, a deep learning model incorporating the full spectrum was constructed for comparative analysis. Ultimately, according to the determination coefficient (R2) of the model, the optimal prediction model was selected for each water quality parameter, with R2 values of 0.756, 0.603, 0.94, and 0.858, respectively. These optimal models were then utilized to visualize the spatial concentration distribution of each water quality parameter within the aquaculture district, and evaluate the rationality of the model prediction by combining manual detection data. The results show that UAV hyperspectral technology can rapidly reverse the spatial distribution map of water quality of aquaculture ponds, realizing rapid and accurate acquisition for the quality of aquaculture water, and providing an effective method for monitoring aquaculture water environments.