Detection of Field Cabbages Using YOLOv8 Deep Learning Technique for an Electric Cabbage Harvester

Detection of fruits and vegetables in actual field conditions is a very difficult task because of various factors such as changing light conditions, occlusion, and shadows. In our study, a self-propelled electric cabbage harvester is to be developed where a vision-based precise cabbage harvesting technique is needed for lifting the cut cabbage with minimum power consumption and minimum damage. For this purpose, the YOLOv8 object detection model was used due to its higher speed of detection. A custom data set of Indian cabbage varieties containing 1289 images was trained on Google Colab. The generated model successfully identified the cabbages with a mean average precision (mAP) of 93.8%, an F1 score of 0.9 at a confidence level of 0.149 during testing and validation. The confusion matric showed that with an average accuracy of 87%, the model detected the field cabbages correctly. The results indicated that the developed cabbage detection model could be used for the precise pushing of cut cabbage in the proposed harvester.