Classification of Potatoes Using Artificial Intelligence Techniques in High Andean Areas of Peru

Agriculture is one of the means that provide food to the population, hence the need to improve its performance levels. Thanks to the artificial intelligence and image processing tools that are currently available, it is possible to improve the planting processes of various agricultural products. In Peru, one of the most produced products, both for domestic consumption and for export, are potatoes, in its various classes, so it is necessary to have a tool that can help farmers in their selection processes. In this research we configured a low-pass and power consumption device, such as model 5 of the Raspberry pi embedded card, where the YoloV3 model is configured, which has the ability to recognize different types of potatoes. For the training of the model, a database of 1791 images was configured, training with 80% of the images and validation of the trained model was performed with the remaining 20%. As a result, the sensitivity of the model was 82%, which means that the model can be used as an aid in classification by farmers. In conclusion, it is indicated that the level of sensitivity can be increased by training with a greater number of images and with the increase of potato types, additionally the classification can be configured in recorded video as in the online recording from a camera, for which performance levels can be selected in terms of the probability of success in the classification.