Medicinal Chrysanthemum Detection under Complex Environments Using the MC-LCNN Model.

Area of Science:

• Agricultural Robotics
• Computer Vision
• Deep Learning

Background:

• Accurate real-time detection of medicinal chrysanthemums is crucial for selective harvesting robots.
• Complex, unstructured field environments pose significant challenges to current detection methods.

Purpose of the Study:

• To develop a novel lightweight convolutional neural network (MC-LCNN) for efficient and accurate medicinal chrysanthemum detection.
• To enhance the model's performance in terms of speed and precision for real-world agricultural applications.

Main Methods:

• Proposed a novel lightweight convolutional neural network (MC-LCNN) incorporating MC-ResNetv1 and MC-ResNetv2 residual structures.
• Integrated custom feature extraction and fusion modules to optimize gradient flow.
• Employed a custom loss function to improve detection precision.
• Embedded the MC-LCNN model into an NVIDIA Jetson TX2 edge computing device using a CPU-GPU multithreaded pipeline.

Main Results:

• Achieved an inference speed of 109.28 FPS (at 416 × 416 resolution) on an NVIDIA Tesla V100 GPU.
• Reached a detection precision (AP50) of 93.06%.
• Improved inference speed by 2 FPS on the NVIDIA Jetson TX2 edge device through multithreaded design.

Conclusions:

• The MC-LCNN model demonstrates high efficiency and accuracy for medicinal chrysanthemum detection in complex environments.
• The model's performance on edge devices makes it suitable for real-time object detection in agricultural robotics.
• MC-LCNN offers a promising foundation for developing advanced perception systems for selective harvesting robots.