基于亮度校正和AdaBoost的苹果缺陷在线识别

On-line Identification of Defect on Apples Using Lightness Correction and AdaBoost Methods

提出了一种基于亮度校正和AdaBoost的苹果缺陷与果梗-花萼在线识别方法。以富士苹果为研究对象,首先在线采集苹果的RGB图像和NIR图像,并分割NIR图像获得苹果二值掩模;其次利用亮度校正算法对R分量图像进行亮度校正,并分割校正图像获得缺陷候选区(果梗、花萼和缺陷);然后以每个候选区域为掩模,随机提取其内部7个像素的信息分别代表所在候选区的特征,将7组特征送入AdaBoost分类器进行分类、投票,并以最终投票结果确定候选区的类别。实验结果表明,该算法检测速度为3个/s,满足分选设备的实时性要求,且总体正确识别率达95.7%。

An algorithm was proposed to on-line identify the defects and stem- calyx on apples based on lightness correction method and AdaBoost classifier. The ＇Fuji＇ apples were selected as the experiment object. First, the RGB images and NIR images of apples were acquired, and NIR images were binarized to obtain the mask images. Second, the R component images were corrected by using proposed lightness correction algorithm and the defect candidate regions were obtained by binarizing the corrected images with a single threshold. Third, every candidate region was treated as a mask, and the information of random seven pixels in the candidate region were selected as the characteristics of the selected candidate region. Finally, an AdaBoost classifier was used to classify these candidate regions by voting way, and the category of candidate region was determined according to the final voting results. For the investigated 140 samples, the results with a 95.7% overall detection rate under acquisition speed of three apples per second indicated that the proposed algorithm was effective.