ARTIFICIAL INTELLIGENCE MODELS BASED ON DEEP LEARNING APPLIED TO THE DETECTION OF OLIVE FRUITS IN TACNA, PERU

Osco Mamani E.F.; Yanapa Chicalla A.D.; Santana Carbajal O.I.; Chaparro Cruz I.N. · Proceedings from the International Congress on Project Management and Engineering (2024)

Resumen

Climate change, El Niño, infectious diseases, foliage and fruit shade can have a significant impact on automated counting and estimation of productivity in olive fruit plots. With an area of 35,000 ha, the Tacna Region is the main producer with 81% of the olives in Peru and is the second largest exporter and third largest producer of olive oil in South America. The objective of this paper is to compare the performance of AI models for olive fruit detection. This work uses a set of 10,728 images obtained from cropping 447 photographs of olive trees in La Yarada Los Palos, Tacna, Peru. This study compares artificial intelligence models based on deep learning with 3 architectures applied to object detection: "YOLOv8s, Faster R-CNN 101 and RetinaNet 101". The images were divided into 8,549 (80 %) for training and 2,179 (20 %) for validation. The results of the olive fruit detection investigation show that Yolov8s models improve in mAP "Mean Average Accuracy" of 0.94759, Faster R-CNN101 of 0.78131 and RetinaNet101 of 0.69626.

Instituciones: Universidad Nacional Jorge Basadre Grohmann