摘要: In ordinary near infrared qualitative identification, maize seed were not covered with seed coating agent. While, in actual agricultural market, maize seeds always should be covered by seed coating agents to resist diseases invasi... 展开 In ordinary near infrared qualitative identification, maize seed were not covered with seed coating agent. While, in actual agricultural market, maize seeds always should be covered by seed coating agents to resist diseases invasion and pests, improve germination rate, and increase yield. The kinds of seed coating are many and varied, and it is hard to determine their components. Therefore it is usually necessary to build identification model by maize seeds without seed coating, and then use the model to recognize seeds with seed coating. The maize seeds coating usually mixed by insecticides, fungicides, fertilizer, plant growth regulators, etc. These components often include hydrogen group organic compounds, which have certain absorption to near infrared spectrum. So the seed coating agent has an interference on near infrared spectroscopy qualitative identification effect. It will reduce the performance of conventional machine learning methods significantly. To reduce the influence caused by seed coating, a method of near infrared spectroscopy qualitative modeling based on deep learning method has been proposed in this paper. Firstly, maize seed spectrum without seed coating agent were used as training set, then a qualitative analysis model is constructed by stack auto encoder algorithm and Softmax classifier. With this deep learning model, maize seeds with seed coating can be identified. The experimental results indicated with the method based on deep learning, maize varietal authenticity recognition rate reduction caused by seed coating is controlled within 3\%. 收起