Selection of Superior Rice Seed Features Using Deep Learning Method

Abstract

Indonesia is a tropical country known as an agricultural country, where 88.57% of the population works in the agricultural sector (BPS Indonesia, 2020). Indonesia is rich in agricultural products such as rice, soybeans, corn, peanuts, cassava and sweet potatoes. Rice (Oryza sativia L) is one of the most dominant food commodities for the people of Indonesia. The carbohydrate content per 100 grams of rice reaches 79.34 grams. The main benefit of rice is as a source of carbohydrates and a source of energy for the body. Seed is one of the factors that play a role as a carrier of technology in advanced agriculture, therefore the seeds used must be of good quality. Farmers tend to equate rice seeds from previous harvests, the rice seed classification process is carried out manually through visual observation and soaking rice seeds in a container filled with water, submerged and floating rice seeds are selected for use, and those that float are discarded. But in reality it still produces less than optimal results, for example rice that is less dense and cracked. This study uses a color moment to be extracted using GLCM (gray level co-occurence matrix) then classified with k-NN to determine the class, then uses the SVM model to display the best hyperplane line to separate the two classes, namely superior and non-superior classes after that system tested with confusion matrix. With a continuous and more intense work process, the research entitled Selection of Superior Rice Seed Features Using Deep Learning Methods. The output of this research leads to a conclusion which rice seeds are superior and which are not superior, aiming to optimize the yield of rice with better quality. The research was successfully carried out using the deep learning method with the highest accuracy of 92.85%.