Research on Visual Recognition of Honeysuckle Based on an Improved YOLO v5 Algorithm

• Published in: 2024 Asia-Pacific Conference on Image Processing, Electronics and Computers (IPEC) pp. 266 - 271
• Main Authors: Shao, Gang, Xiong, Haoyu, Qin, Jianshuang, Li, Guimin, Jiang, Yulian
• Format: Conference Proceeding
• Language: English
• Published: IEEE 12.04.2024
• Subjects: Accuracy; Attention mechanisms; CA attention mechanism; data set construction; honeysuckle; Image preprocessing; Machine learning algorithms; Object segmentation; Robustness; Training; Visualization; Web and internet services; YOLO; YOLO v5 algorithm
• DOI: 10.1109/IPEC61310.2024.00053

With the continuous evolution of machine vision technology, its application in agriculture, forestry, and other fields has become one of the focal points of research. This paper proposes an improved model based on machine learning specifically for the visual recognition of honeysuckle. Firstly, by constructing and optimizing the dataset, and utilizing Baidu PaddlePaddle BML for training set creation and fine-tuning, the model is enabled to better learn and understand the characteristics of honeysuckle. Secondly, advanced image preprocessing techniques, including bilateral filtering and the RGB color space model, are employed to enhance the segmentation precision and accuracy of honeysuckle. Subsequently, an innovative Coordinate Attention (CA) mechanism is introduced, allowing the model to focus more on pixel-level important information, thereby improving the recognition performance of honeysuckle. Experimental results demonstrate that the proposed algorithm achieves remarkable performance in the task of honeysuckle recognition, with an average precision of \mathbf{7 8 . 6 \%}, representing a \mathbf{1 0 . 1 \%} improvement compared to the unmodified algorithm. This ensures high-speed recognition based on YOLO v5 while effectively enhancing the algorithm's robustness.