Detection network for multi-size and multi-target tea bud leaves in the field of view via improved YOLOv7

In the natural tea plantation environment, the accurate detection of multi-size and multi-target tea bud leaves within a wide field of view is essential for successful tea picking. However, the detection task is challenging due to factors such as the targets have a high resemblance to the background color, and the size of tea varies across different varieties and growth conditions. Additionally, there are numerous targets in the field of view, all of which contribute to the increased difficulty in detecting tea bud leaves. To address these challenges, this paper presents a novel method for the detection of tea bud leaves. The method incorporates a selective kernel attention mechanism in the Backbone network to enhance the ability to extract morphological features. Then a new multi-feature fusion module is introduced to combine different local features and integrate them with global features, capturing both local and global dependencies, and enabling comprehensive and distinct feature representation. Furthermore, an effective loss function is employed to calculate the loss values for class probability and objective score, penalizing false detections and missed detections during the training process. The experimental results demonstrate that the proposed model improved YOLOv7 achieves superior detection performance and robustness, with a recall rate of 84.95%, precision of 90.99%, and average precision of 94.43%. These values are approximately 10% higher compared to the original YOLOv7 model. The detection network can accurate detection of tea bud leaves in tea plantation environments.