Share
Export Citation
DETECTION AND SEQUENCE OF COLLECTING LOOSE OIL PALM FRUIT USING DEEP LEARNING TECHNIQUE
Kakisina S.E.
Iet Conference Proceedings
Q4Abstract
This paper uses YOLOv7 to detect and collect loose palm fruits with an autonomous harvesting robot. YOLOv7 was chosen due to its high accuracy and real-time detection capability. Data from high-resolution cameras in oil palm plantations are annotated to create a robust training set. Key hyperparameters such as Anchor, Learning Rate, and Weight Decay are fine-tuned to see which performs well in object detection. The robot uses YOLOv7 to detect fruits, prioritizing picking based on the coordinates of the bounding box. Hyperparameter set A excels in detection accuracy and precision, while set B is slightly better in bounding box prediction. MAP50 is 0.866 and 0.823 for sets A and B, respectively. The optimized model significantly improves the efficiency of the robot in real-time harvesting.