CRipS: Coconut Ripeness Stage Detection System

Manual Coconut harvesting requires tree-climbing skill, it is hazardous because of the tree's hard shell, height and high occlusions on the tree. Further, it is difficult to determine whether a coconut is ripe or not because of the outside illumination and various background. The development of a coconut ripen stages detection system with the help of deep learning techniques is of great interest to farmers. This research work proposes a deep learning-based instance segmentation technique to identify different coconut ripeness stages. To improve instance segmentation accuracy to identify different coconut ripeness stages, the SOLOv2 model is used. In this research work, collected image dataset consisting of approximately 500 RGB images of different types of coconut fruits in coconut plantations during daylight under natural lighting conditions using a handheld digital single-lens reflex camera. The proposed SOLOv2 model was trained, validated, and tested on 500 manually acquired and augmented images. On the test dataset, the mean average precision (mAP) and Average Precision (at different intersections-over-union (IoU) thresholds), Average Precision (at different intersection-over-union (IoU) thresholds) attained by the model were 0.628, 0.628 and 0.811, respectively, with an average precision and recall for identifying ripeness stages as 0.912 and 0.883, respectively. I data enables the creation of a comprehensive vision system to pinpoint the cutting spot on the coconut cluster and select the harvesting method. The experimental results enable us to develop an automated ripeness detection system to determine the ripeness stages for harvesting and pinpointing the cutting spot on the coconut tree.