Design of a Corn Type Recognition System Using YOLOv3 Architecture

Computational techniques are allowing a high degree of impact in many areas, one of them is agriculture, where many solutions are being presented, from precision agriculture to the use of artificial vision to analyze different aspects of plants. In this paper we developed a recognition system for the following types of corn: purple, choclo, cancha serrana, Chullpi and Gigante del Cuzco, by means of video analysis using the Yolo V3 model with reinforcement training. For the training process, a database with 100 images corresponding to the indicated types of corn was constructed, separated into two groups, a training group with 80% of the images and a test group with the remaining 20%. The results are presented based on a percentage of recognition, which was performed classifications with 85, 90, 95 and 100% probability of recognition, analyzing the results the probability value of 90% presents a greater amount of positive recognition, compared to the others, calculating a level of classifier performance at 92%. As a conclusion, we indicate the scalability of the proposal, to increase the amount of corn types, as well as to increase the number of images in the database, to cover a wider spectrum within the types of corn that exist in Peru.