The Research Of On-line Biological Early Warning Technique For Water Quality Based On Fish Tracking And Behavior Analysis

With the rapid development of social economy, the environmental pollution has been moreserious，which restricts the economic development and begins to attract wide attention. Thefrequent occurrence of the water pollution accidents poses a serious threat to the balance of theecological environment and human life safety. Therefore, many researchers have focused on howto monitor water quality accurately and quickly in the recent years. The methods of detecting waterquality are physicochemical analysis method and biological monitoring method. Thebio-monitoring method combined with computer vision technology can reflect the water qualityquickly, accurately and conveniently.This paper, using computer vision technology, focuses on the early warning technique basedon fish tracking and behavior analysis. In the experimental process, the zebrafish was selected asthe test organisms. And copper ion solution was used as the toxic reagents. The details of theresearch work are as follows:1. An experimental system was designed, which can monitor the fish shoal day and nightcontinuously. The platform can work in day mode and night mode for observing the fish behavioraffected by circadian variation. In addition, the experimental platform can simulate the acutepoisoning process of the fish shoal and monitor the changes of the fish behavior in the poisoningprocess. Eventually, it makes early warning for water quality.2. In this paper the computer vision techniques were used to obtain video images of fishbehavior and get the fish targets through background model algorithm. Then the group parametersof the fish were quantified, including the average speed, the distribution and the intensity of the fishshoal, the amount of fish on the surface and on the bottom. The results show that backgroundmodel method can identify the fish quickly and accurately under the experimental conditions, andthe parameters can be obtained in real time.3. When the water quality changes, the swimming trajectory of individual fish will alsochange. This paper established a matching matrix between two continuous frames and combined the swimming direction and posture of the fish to obtain the motion trajectories. Experimentalresults show that when the fish is in the single target, the target fusion, the target separation and thetarget rupture statuses this algorithm can obtain ideal tracking results. Then, according to thetrajectory, the individual characteristic parameters of the fish shoal were quantified, including thefractal dimension, the speed parameter and the direction parameter.4. A difference analysis of the group parameter data was made during the day mode and nightmode before and after fish poisoning. The results show that whether during the day or night, whenthe water quality changes, the group parameters have significant differences. On the other hand, byanalyzing the individual quantization parameter, we found that the complexity of the trajectories,the velocity magnitude of the fish and the polymorphism of the speed direction are all quitedifferent before and after the treatment.5. Adaboost algorithm was used to solve classification problem of the quantization data fromthe normal and abnormal water in the day mode and the night mode respectively. To evaluate theaccuracy and the time consumed of the classification, Adaboost algorithm was compared tosupport vector machines (SVM) and back propagation neural network (BPNN). The results showthat the accuracy of Real Adaboost and Gentle Adaboost algorithm was greater than93%, whetherin the day mode or in the night mode. The results were better than the Modest Adaboost, BPNNand SVM, and the time consumed was fewer. The results of this paper prove that Adaboostalgorithm is an effective method for the classification of the water quality parameters. We can usethis method for water early warning, to achieve the online warning automatically.