A lightweight trichosanthes kirilowii maxim detection algorithm in complex mountain environments based on improved...

Area of Science:

• Computer Vision
• Agricultural Robotics
• Machine Learning

Background:

• Automated harvesting of Trichosanthes Kirilowii (Cucurbitaceae) in complex mountain environments is hindered by environmental challenges like varying brightness, occlusion, and motion blur.
• Existing detection algorithms often suffer from excessive parameters and high computational intensity, limiting their practical application in resource-constrained harvesting robots.

Purpose of the Study:

• To propose a lightweight and efficient detection algorithm, KPD-YOLOv7-GD, for accurately identifying Trichosanthes Kirilowii in challenging mountainous terrains.
• To improve the feature extraction efficiency and detection accuracy of YOLOv7-tiny for real-time agricultural robotic applications.

Main Methods:

• The study adapted YOLOv7-tiny, enhancing its multi-scale feature layer and incorporating a lightweight convolutional module and channel pruning to reduce model complexity.
• Integration of the Dynamic Head (DyHead) module and Content-Aware Re-Assembly of Features (CARAFE) module, alongside knowledge distillation techniques, further optimized feature extraction and detection performance.

Main Results:

• The KPD-YOLOv7-GD algorithm achieved a mean average precision (mAP) of 93.2% for Trichosanthes Kirilowii detection.
• Compared to mainstream single-stage algorithms, KPD-YOLOv7-GD demonstrated significant mAP improvements (e.g., 4.8% over YOLOv3-tiny) with substantial model compression rates (e.g., 81.6% for YOLOv3-tiny).

Conclusions:

• KPD-YOLOv7-GD offers a superior balance of low complexity, high recognition accuracy, and speed, making it ideal for resource-constrained automated harvesting robots.
• The developed algorithm effectively addresses the challenges of detecting Trichosanthes Kirilowii in complex environments, paving the way for more efficient agricultural automation.