基于多光谱图像的库尔勒香梨表面缺陷检测

Surface Defect Detection of Korla Pear Based on Multi-spectral Image

表面缺陷会导致水果腐烂,降低商业经济价值,引发食品质量和安全问题。为此,利用机器视觉技术检测果皮表面缺陷是水果采后加工的主要任务之一。然而,由于香梨果皮颜色及缺陷多样性,用图像处理方法检测库尔勒香梨表面缺陷较为困难。为实现快而准确的香梨缺陷检测,提出一种基于多光谱图像的库尔勒香梨表面缺陷检测方法,并研究了疤痕、病害、虫咬、碰压伤、机械损伤及果锈6种表面缺陷情况。同时,设计了一种基于多光谱图像的库尔勒香梨表面缺陷面积计算方法,通过统计缺陷区域像素个数来代替缺陷面积。按照国标GB/T19859-2005规定:选定缺陷面积大于0.8cm 2作为缺陷果的判别依据进行试验,结果表明:带有疤痕、病斑和腐烂缺陷的缺陷识别准确率可达到92%以上,处理单张图像平均耗时2.4s,具有较高准确性,可用于库尔勒香梨的实时检测。

Surface defects can lead to fruit decay,reduce commercial value,and cause food quality and safety problems.Therefore,using machine vision technology to detect the surface defects of fruit skins is one of the main tasks of post-harvest processing.However,it is difficult to detect the surface defects of korla pear by image processing because of the variety of skin color and defects.In order to realize fast and accurate defect detection of korla pear,a new method based on multispectral image was proposed.Scar,disease,insect bite,crush,mechanical damage and fruit rust were studied.A multispectral image based method for calculating the surface defect area of korla fragrant pear was proposed.According to the national standard GB/t19859-2005,the selected defect area is larger than 0.8cm 2 as the discriminating basis of the defect.The test results showed that the accuracy of defect recognition with scar,diseased spot and decay could reach more than 92%,and the average time of processing single image was 2.4s,with high accuracy.It can be used for real-time detection of korla fragrant pear.