Evolutionary Feature Optimization and Classification for Monitoring Floating Objects

摘要

Water surfaces are polluted due to various man-made and natural pollutants. In urban areas, natural water sources including rivers, lakes and creeks are the biggest collectors of such contaminants. Monitoring of water sources can help to investigate many of details relating to the types of litter and their origin. Usually two principle methods are applied for this type of applications, which include either a use of in-situ sensors or monitoring by computer vision methods. Sensory approach can detect detailed properties of a water including salinity and chemical composition. Whereas, a camera based detection helps to monitor visible substances like floating or immersed objects in a transparent water. Current computer vision systems require an application specific computational models to address a variability introduced due to the environmental fluctuations. Hence, a computer vision algorithm is proposed to detect and classify floating objects in various environmental irregularities. This method uses an evolutionary algorithmic principles to learn inconsistencies in the patterns by using a historical data of river pollution. A proof of the concept is built and validated using a real life data of pollutants. The experimental results clearly indicate the advantages of proposed scheme over the other benchmark methods used for addressing the similar problem.