Feature Extraction in the Image Recognition for Plateau Pikas (Ochotona curzoniae)

Abstract

The conventional methods for pika observation and investigation are costly both in material resources and manpower, moreover these approaches are difficult to use in long-term and continuous investigation. In this context, a plateau pikas monitoring scheme based on image processing technology was described, and the steps of pika identification is presented. Furthermore, the most challenging issue of pika identification, feature extraction was analyzed in detail. Based on the analysis, the image graying method was utilized to avoid redundance and compress data capacity, and image binarization technology was utilized to eliminate the background noise. Moreover, in order to effectively extract the features, four edge detectors: Sobel, Prewitt, Roberts and Canny were chosen for evaluation, and among these algorithms Roberts detector were found has better performance in the application. At last, median filtering method was utilized in image smoothing for spiky noise elimination and sharp edges preservation. The experiment results showed that the pika detection method based on image processing has the advantages to achieve dynamic and continuous monitor of pika in close quarters, and significantly improved the technique in meadow biogeocenose protection.