Onions classification automation using deep learning with convolutional neural network method

TLDR: A method for classifying the Convolutional Neural Network (CNN) method is a development of Deep Learning techniques in terms of object recognition and object classification in high resolution images.

Abstract: Earn a living as farmers, supported by the presence of fertile soil and a tropical climate that is suitable for use in the agricultural sector, for example the horticulture subsector. Onion is an agricultural commodity that is needed by all people around the world. Onions are used as food flavoring enhancers and have great efficacy in terms of treatment. Onion needs always increase every year in accordance with high market demand which is in line with an increase in population, while onion production is seasonal [1]. In the district of Magelang, several farmers planted onions by mixing all types of onion seeds in one field. This is due to the unavailability of farmers’ land to be planted. After the harvest, all onion production is made into one place so that it is easier to package and distribute to onion suppliers. This makes the supplier takes a long time in separating the types of onions. Therefore, we need a technology and system to facilitate the filtering of onions based on the type so that, it is easy to recognize the species quickly and automatically with a more efficient time. With the Deep Learning technique, the machine is expected to be able to classify the differences between the onions. One method for classifying the Convolutional Neural Network (CNN) method is a development of Deep Learning techniques in terms of object recognition and object classification in high resolution images. Based on the results of the analysis conducted, the best architecture was obtained using a 80% data train comparison scenario; test data 20%, filter size 10, kernel size 3x3, size of learning rate 0.01 using ReLu activation, number of epoch 70, batch size 50 which uses the type of color image (RGB). produces an accuracy of 70%. Where for the results of the classification of predicted images of onions according to its category as many as 8 images for onions, 9 images for onions, and 4 for garlic.