Classification of good and damaged rice using convolutional neural network

Abstract

Rice production is massive in Indonesia, therefore maintaining the quality of the product is necessary. Detection and classification of objects have become a very important part in image processing. We performed object detection namely rice. After the object is found, it can be classified into two categories, namely good and damaged rice. We conducted a new study on rice which was carried out per group not per grain to obtain or classify good and damaged rice where we had carried out several steps, namely segmentation process using HSV (hue, saturation, value) color space. HSV is used because of its excellence in representing brightness of the image. We considered evaluating brightness because the tendency of damaged rice is darker or paler compared to good rice. To accomodate environment lighting ambiguity we perform the image acquisition in a controlled environment, so that all the images have the same light intensity. Here we use only channel V of HSV to be used in feature extraction using the gray-level co-occurrence matrix (GLCM) and finally convolutional neural network (CNN) is used for classification. From the test experiments that we have done, we have produced 83% prediction accuracy. Considering how similar the good rice is to the spoiled rice, the results are quite impressive.