Abstract: 【Aim】 Among many species of oil-tea camellia pests, the tea tussock moth, Euproctis pseudoconspersa, is one of the most dangerous pests. Aiming to complete the automatic detection of E. pseudoconspersa larvae, the images of the oil-tea camellia pests need to be segmented, and the segmentation effects influence directly the automatic identification of images. 【Methods】 In this paper we proposed an new algorithm based on K-means and region merging to make difference calculation for three components of RGB. The originally segmented images were gotten through combining adjacent pixels if the maximum difference is equal to 0. According to the merging rules, the pixels were further merged to obtain the final segmentation results. 【Results】 The proposed algorithm in this paper could separate image background from pests quickly and effectively. 【Conclusion】 Through comparsion of the image segmentation effects from JSEG argorithm, K-means segmentation argorithm, fast geometry deformable segmentation and the proposed algorithm, we found that the proposed algorithm is better than others because it could get satisfactory results within very short time.

Key words: Euproctis pseudoconspersa, automatic identification, region merging, maximum neighborhood difference, image segmentation