Estimating the aquatic-plant area on a pond surface using a hue-saturation-component combination and an improved Otsu method

Abstract

Accurately estimating the aquatic-plant area on a pond surface is of great significance for regulating nutrients, such as nitrogen and phosphorus, increasing the dissolved-oxygen content, and ensuring the healthy growth of aquaculture organisms. Currently, remote-sensing methods for monitoring pond-surface aquatic plants have a high cost, long cycle, and poor real-time performance, which limits the practicability of these methods in aquaculture. To address the above problems, a pond-surface aquatic-plant area estimation method based on a hue-saturation (H-S)-component combination and an improved Otsu method, was proposed, which utilizes the advantages of fast and non-destructive computer-vision technology. First, the pond-surface images collected by the camera were preprocessed using a median filter. Second, a visual-saliency map of the aquatic plants was generated by combining the H and S components of the HSV color space. Then, the appropriate Otsu optimal threshold-selection criterion formula was selected to determine the optimal segmentation thresholds of the aquatic plants and background water. Finally, the aquatic plants in the image were segmented using the binary method, and the aquatic-plant area on the pond surface was estimated, based on the proportion of aquatic-plant pixels. In this study, an improved Otsu method was proposed to adaptively select the Otsu optimal threshold-selection criterion formula by judging the position of the highest and second highest peaks of the gray histogram of the visual-saliency map, which solved the problem of finding the appropriate segmentation threshold for different proportions of aquatic plants and background water. The proposed method was tested on real pond-surface images with a mean misclassification error of 0.0284, mean F1 score of 0.8396, mean precision of 0.8428, and mean recall of 0.9146. The experimental results indicated that the proposed method could accurately segment aquatic plants to accurately estimate the aquatic-plant area on the pond surface.