The thesis focuses on the automation of agricultural processes using advanced computer vision and artificial intelligence systems, particularly in asparagus harvesting. Due to the growing global population and limited agricultural land, automation is becoming crucial for ensuring sufficient food supply. The work provides a comprehensive analysis of modern approaches to crop harvesting automation, including robotic systems for apples, tomatoes, and weed detection, which are based on mobile platforms, depth cameras, and algorithms such as YOLO and PointNet.
Special attention is given to harvesting stem vegetables, where detection and localization are key. Various approaches are presented, ranging from tactile sensors and laser scanners to depth cameras, with the RGB-D depth camera emerging as the optimal choice. The implementation of asparagus detection is based on the YOLOv8 algorithm, enabling real-time object detection and the processing of heterogeneous scenes with high accuracy. The model was trained on annotated images of asparagus models and real asparagus, achieving satisfactory results, although a larger dataset and further optimization are still required for real-world applications.
Additionally, the use of point clouds is presented for analyzing the three-dimensional properties of asparagus, such as height, position, and shape, allowing for quantitative evaluation for automated harvesting systems. The importance of high-quality data, proper calibration, and optimized acquisition is emphasized, ensuring the accuracy and reliability of the system.
|
