基于近红外高光谱图像的苹果轻微损伤检测

Detection of slight bruises on apples using near-infrared hyperspectral image

针对苹果轻微损伤时,基于可见光的机器视觉方法难以有效检测的缺点,开展了近红外高光谱图像的苹果轻微损伤检测研究.首先,用900-1 700 nm近红外波段范围对轻微损伤苹果高光谱成像,图像显示损伤部分与正常部分区别明显.其次,采用特征波段比方法和不均匀二次差分方法对损伤苹果光谱图像进行处理,增强损伤处与正常位置的可分性.最后,利用3种分割方案,对损伤部分进行自动分割.对50个含轻微损伤和正常的苹果进行分割,实验结果表明,不均匀二次差分方法的损伤检测准确率为92%,比主成分分析法和波段比方法具有更高的检测准确率,为轻微损伤苹果的准确检测提供了一种新的方法.

A research of apple slight bruises was conducted by using hyperspectral images,aimed at solving the difficulty of the traditional defect detection method of machine vision. This study is in part based on the fact that visible light faces great challenges on it. First,the hyperspectral images of slight bruise apples between 900 and1 700 nm are acquired by a hyperspectral imaging system. It can be found that the differences between the normal part and the bruise part are obvious. Next,we analyzed the hyperspectral images via the feature band ratio method and asymmetric second difference method to improve the divisibility of the normal part and the bruise part. Finally,the bruise parts were automatically segmented from the normal part with three defect detection methods. The experimental results show that the accuracy of detecting slight bruises on the 50 apples using asymmetric second difference method is 92%,which is higher than the principal component analysis and band ratio methods. Therefore,the work provides a new method to detect the slight bruise apples accurately.