Authors: Yang Yan, I.V. Chicherin, Zhao Lijun, Long Chanjuan, E.A. Ignatieva

Title of the article: Development of recognition algorithms in the control system of robot drives for plucking tea

Year: 2025, Issue: 2, Pages: 12-21

Branch of knowledge: 2.4.2. Electrotechnical complexes and systems (engineering)

Index UDK: 62-5:004.5

DOI: 10.26730/1816-4528-2025-2-12-21

Abstract: This study focuses on image recognition of tea Shoots using an object detection algorithm based on a deep learning framework. The developed algorithms are used in the control system of the robot's drives for collecting tea. The morphological requirements for tea shoots are one bud with one leaf or one bud with two leaves. This paper uses the YOLOv5 object detection algorithm to establish a recognition model for tea shoots. The reliability of the trained model is evaluated using four training indicators: precision, recall, F1 score (harmonic mean of precision and recall), and mean average precision (mAP). Firstly, image enhancement is performed, including converting the images to grayscale, sharpening, denoising, and symmetric mirroring. The enhanced images are manually annotated using the Labeling tool and exported in YOLO format as .txt files, resulting in a folder containing the labeled images. Secondly, The training is conducted in Pycharm with the processed images divided into batches, and the respective performance indicators are obtained. Finally, all the candidate boxes with confidence greater than 0.65 are retained without any missed detections, The grayscale conversion makes the processing effect more obvious when extracting feature maps during the YOLOv5 training process, because grayscale images only have black and white gradients. This ultimately leads to better training results. However, factors such as curled leaves, residual leaves, color, lighting, shooting environment, and shooting angle, which are characteristic of tea leaves in reality, can lead to decreased composite evaluation metrics and thereby affect the experimental results to some extent.

Key words: object detection recognition algorithms drive control YOLOv5 tea shoots

Receiving date: 25.03.2025

Approval date: 01.05.2025

Publication date: 05.06.2025

CITE DOWNLOAD