Study On Anomaly Water Quality Assessment Factor Based On Computer Vision

Water quality environmental security has become one of the important constraints of China’s social and economic sustainable development. In recent years, many approaches have been proposed to evaluate water quality by using established indicators, most of which are collected from physical and chemical analysis of polluted water. However it will lead to other evaluation issues, such as subjectivity, and this may not truly reflect water quality. In order to protect water quality environmental safety, study of anomaly water quality assessment factor based on computer vision was proposed. The paper researches into how to select the general, comprehensive and representative factors which reflects the main relationship between water environment and fish behavior. The main contents of this paper were shown as follows:Firstly, the paper mainly focuses on using red crucian carp as an indicator of water pollution. Fish movement information is collected through computer vision and while to track, analysis and forecast the changes of fish behavior. The fish motion video sequences pretreatment system is built based Bilateral filtering and discrete cosine transform(DCT),which is better to detect and track the red crucian carp.Secondly, anomaly water quality monitoring programme has been designed based computer vision and support vector machine(SVM). Kernel function type and parameter optimization have a significant impact on the anomaly water quality monitoring model. So by comparing the different types of kernel function experimental results to choose the best kernel, then using particle swarm optimization algorithm(PSO) and genetic algorithm(GA) and the Grid Search method to optimize parameter.Finally, by searching the effect between water quality change and the trajectory of red crucian carp, the curvature and vicinity feature calculated by fish trajectory are introduced.Then velocity, acceleration, curvature and vicinity feature are extracted from red crucian carp under normal water quality and water added dichlorvos, respectively, then feature databases are established for water quality detection. The classification is conducted by support vector machine using evaluation indexes, and by the analysis of ROC curve, theaccurate rate of anomaly evaluation on relationship between water environment and fish movement behavior is investigated. The comparison experimental results show that curvature and vicinity feature as water quality assessment factor is obviously better than velocity and acceleration, especially abnormal water quality recognition rate is over 90%.Moreover, the ROC curve has proved models based curvature and vicinity feature are steady, universal and effective.