Multi-scale edge detection of rice internal damage based on computer vision

Abstract

Damage is easy to occur during the process of harvesting, classification, packaging, transporting, processing, preservation and selling etc. and the detection of internal damage is a difficult problem. Rice kernels were classified as those with none, single, double or multiple stress cracks. An image processing algorithm was used to enhance the object and reduce noise in the acquired image. At the same time a machine -vision system was developed to detect different types of stress cracks in rice kernels. None and single stress cracks were the easiest to detect. Careful positioning of the kernel over the lighting aperture was necessary for accurate detection of double and multiple stress cracks. This system provided an average accuracy of approximately 96.5% to none crack, 93.4% to single crack, 84.2%to double cracks and 83.4% to multiple cracks compared to human inspection. The processing time was between 0.45 and 0.12 s/kernel.