Water Quality Anomaly Monitoring Based on Kalman Filter and Convolution Neural Network

Abstract

Aiming at the problem of integrated environmental monitoring of water quality pollution, a method of water quality anomaly monitoring using Kalman filter and Convolution Neural Network(CNN) is proposed. In this method, the fish is segmented by Mask R-CNN image segmentation method, the positive and negative sample data sets of the backbone and background images of the fish are made, and the model is obtained by using convolution neural network training data set. In the process of tracking, the RANSAC algorithm is used to screen the SIFT feature matching and Kalman filter is used to track the fish and draw the moving track in real time. The motion trajectory is saved every 3 seconds, a total of 150000 samples of normal and abnormal water quality are obtained. The experimental results show that the recognition rate of water quality anomaly based on Kalman filter and Convolution Neural Network is 98.5%, this method is superior to traditional water quality identification methods.