Detection of on Tree Maturity Stage of Guava Fruit Using Object Detection Model

Research explores the application of advanced object detection techniques to assess the ripening stages of guava fruits on trees using the YOLO model. The various stages of fruit – unripen, ripen and defective are considered. The dataset consists of images with varying illumination and intensity to include diverse data. Also, the augmentation techniques are used to increase the data size. The Yolo model is used. The various performance such as box loss, class loss and object loss is measured. The multiple graphs such as precision confidence, recall confidence, and F1 curve are represented. To provide detailed class wise evaluation, the confusion matrix is used and the ROC curve is determined. The YOLOv5 model is compared against the traditional machine learning models – SVM, Random Forest and Decision Tree. The mAP is also measured. The Cohen’s Kappa co-efficient is evaluated and shows value of 0.92 that demonstrate the efficacy of the model in automating quality assessment and harvesting decision making, providing a robust tool for precision agriculture.