关键词: 双孢菇, 改进RHT, 均值漂移聚类, 目标提取

Abstract: In the factory production environment, Agaricus bisporus images have complex states such as cluster adhesion, tilted fungiculum and abundant white mycelia, which cannot be accurately extracted by conventional threshold segmentation method. Under such background, this paper combined an improved Random Hough Transform (RHT) and randomized circle detection algorithm (RCD) with the moving average clustering algorithm (MVCA) for target extraction. Some background interference was reduced through image preprocessing. The sampling method at the edge points was improved to increase the circle detection efficiency. The invalid parameters were eliminated based on the radius of the pileus. All the circles detected by the Agaricus bisporus images were taken as the target. Then MVCA was used to merge the same circles and eliminate erroneous targets identified in the circle detection algorithm .The algorithm was used to extract the target results. The accuracy of target extraction was over 92% for images of Agaricus bisporus with regular growth, pileus covered by soil, low overlapping ratio or small deformation of pileus. While for images of Agaricus bisporus surrounded by largearea hyphae, tilting in clusters, uneven in size or with serious pileus piling up and deformation, the target extraction accuracy was more than 82%, and there were problems of inaccurate circle detection, misjudgment of clustering or missed judgment. With the advantages of small amount of algorithm data, fast calculation and strong adaptability, the RHT- RCD-MVCA integrated method proposed in this paper could meet the needs of automatic monitoring and fruiting statistics during the growth of the Agaricus bisporus.

Key words: Agaricus bisporus, improved RHT, MVCA, target extraction