USING YOLOV5 TO DETECT PLASTIC GREENHOUSES IN MOROCCO

Abstract

Globally, Increasing greenhouse production is causing producing environmental issues. Monitoring the number and determining the locations of greenhouses are essential for the modern agricultural land management. The aim of this study was to detect plastic greenhouses using remote sensing data and the YOLOv5 CNN-based object detection algorithm. In the process, 426 images of plastic greenhouses were manually downloaded from Google and Mohammed 6 satellites and were divided into three datasets, training (63%), validation (15%), and testing (22%). Plastic greenhouse images in the training and validation sets were manually labeled, with the training set being used to train the YOLO-V5 Model. Throughout the training, the validation dataset was used to evaluate how well the network performed. According to the performance results the system reached its highest level of accuracy at epoch 71, with a relatively brief training time of 40 minutes, The model’s average precision was 84.1 percent, Using satellite images. In detecting plastic greenhouses, YOLOv5 can deliver exceptional accuracy and computational performance.