基于计算机视觉的柑橘糖度检测  

Determination of sugar content of citrus fruit based on computer vision

作　　者：曹乐平[1] 黄中培[1] 林继元[1] 

机构地区：[1]湖南生物机电职业技术学院,长沙410127

出　　处：《食品科技》2008年第6期186-188,共3页Food Science and Technology

基　　金：湖南省农业厅科学研究项目(200704);湖南省教育厅科学研究项目(06D059)

摘　　要：现场采摘180个涟红温州蜜柑,考虑色泽和大小范围的广度,从中选取140个作为试验样本,各果转90°中心角,采集一幅图像,每果采图4幅,通过图像裁切,RGB空间至HSI空间的转换和图像差值法去背景,提取柑橘色调H和饱和度S表面色泽参数,用110个样本训练小波神经网络,30个样本检验网络性能。试验结果表明,检测最大绝对偏差0.2452°Brix,最小绝对偏差0.0002°Brix,平均偏差0.0545°Brix,标准差0.0830°Brix,精度在±0.1°Brix内的正确识别率为73.33%,精度在±0.2°Brix内的正确识别率为90%。This paper investigates method of determination of sugar content of citrus fruit based on computer vision. The testing samples are 140 Lian-Hong Wenzhou citrus fruits which cover wide range over size, shape and color. Four images are taken from each fruit in every 90° central angle. Images are cut to normal size and converted from RGB to HSI space, and removed background with image subtraction. Hue and saturation of each image are extracted. A wavelet neural network model is proposed and trained with 110 samples. Model performance is tested with other 30 samples. The test results show： the maximal absolute deviation is 0.2452 °Brix, minimal absolute deviation is 0.0545 °Brix, standard deviation is 0.0830 °Brix. The correctness of detection for accuracy ±0.1 °Brix and ±0.2 °Brix are 73.33% and 90% respectively.

关 键 词：柑橘 糖度 计算机视觉 小波神经网络 图像处理 

分 类 号：TS207.3[轻工技术与工程—食品科学]