分选加工中稻米垩白自动检测算法

Study on rice chalkiness automatic detection algorithm for sorting processing

垩白米不仅影响稻米的外观品质,还影响米粒的食味品质,降低稻米商品价值。该文以稻米加工中不同稻米籽粒的组合图像为研究对象,给出了基于切比雪夫逼近的垩白自动分割算法和稻米垩白指标测定算法。算法可实现自适应阈值选取,对包含黄米和杂质米的组合米样图像也能实现垩白的完整分割,算法鲁棒性强。按照国家标准要求,选取垩白粒率为40%的100粒稻米进行随机组合抽取来验证算法的准确性和实时性,结果显示垩白粒率检测的准确度为95%,垩白度的计算误差为2.39%,稻米垩白检测平均耗时3.8 ms/粒,算法耗时短适合在线运算。该文算法用于稻米加工中垩白米的分选,可提高加工后稻米的商品价值和食味品质。

The rice chalky portion is defined as the opaque white portion in rice endosperm. Chalky rice not only affects its appearance quality, but also affects its cooking and taste quality, and then reduces the rice commodity price. Therefore, picking chalky grain in the processing of rice sorting has important practical value and economic value. In this paper, different rice combination images appearing in the sorting process was researched, and the rice kernels＇ chalky portions were segmented automatically using image processing technology. According to the national standard requirements, chalky degree and chalky rice rate as rice chalky indexes were determined. First, the background image of the multi-grain rice image was segmented automatically in I color channel using an Otsu algorithm. Then, the segmented binary image and the original image were phased to get the rice image while removing the background. Viewing the rice transparent part as background and the rice chalky part as the foreground, the image was automatically segmented again using a Chebyshev approximation algorithm. The fake chalky areas in the image were removed using the area threshold method in a twice segmentation process. In this paper, a rice chalky portion automatic recognition algorithm and a chalky rice index detection algorithm were given and experimentally analyzed from their robustness, accuracy, and time-consuming aspects. The results showed that the algorithm could implement adaptive threshold selection, and realize the chalkiness complete segmentation of a combination image especially an image including yellow rice and rice with impurities, so the algorithm robustness was strong. According to the national standard requirements, one hundred rice kernels with 40% chalky rice rate were selected and different rice kernel images with a random combination were segmented to verify the accuracy and time-consuming of the algorithm. The results were that the chalky rice rate accuracy was 95% and the calculation error of the chalky degree was 2.39%. The chalkiness detection average time of each rice kernel was 3.8 ms, and the algorithm counting time was short and suitable for online operations.