Abstract: The production of silk reeling and weaving in our country has already entered a stage of mechanization and automation. However, cocoon image recognition technology is still in its early stages of research, and there are many technical bottlenecks in its promotion and application. In order to better apply the research of intelligent recognition technology to cocoon sorting work and promote the healthy and efficient development of the silk industry, this article starts from the classification of cocoons and the standards for selecting cocoons, and introduces the current research status and characteristics of cocoon image recognition technology based on color features, shape features and texture features. It summarizes and analyzes the problems of incomplete recognition information, limited recognition types, low recognition accuracy and efficiency in current cocoon image recognition technology research. Suggestions are put forward for conducting research on the establishment of complete silkworm cocoon images, increasing the collection and discrimination of multiple features, and integrating deep learning technology. The development trend of intelligent and efficient silkworm cocoon sorting is also prospected.

Key words: cocoon sorting, image recognition, feature recognition, machine vision, deep learning

摘要： 我国缫丝织绸早已进入机械化、自动化的发展阶段，但是蚕茧图像识别技术仍处于研究的初级阶段，在推广和应用中存在很多技术瓶颈。为更好地将智能识别技术的研究用于蚕茧分选工作，推动蚕丝产业的健康高效发展，从蚕茧的分类和选茧的标准出发，分别介绍目前蚕茧图像识别技术研究中基于颜色特征、形状特征和纹理特征的识别技术研究现状和特点，总结和分析现在蚕茧图像识别技术研究中存在识别信息不完整、识别种类少、识别准确度和效率较低等方面的问题。提出在开展完整蚕茧图像建立技术、增加多种特征的采集和判别、开展深度学习技术融合的研究建议，展望蚕茧智慧高效分选的发展趋势。

关键词: 蚕茧分选, 图像采集, 特征识别, 机器视觉, 深度学习