Video based oil palm ripeness detection model using deep learning.

Area of Science:

• Computer Vision
• Agricultural Technology
• Machine Learning

Background:

• Existing oil palm detection research often uses static images, limiting real-time application.
• Real-time detection is crucial for automated harvesting and increased efficiency in oil palm plantations.

Purpose of the Study:

• To develop an object detection model optimized for real-time oil palm detection using video datasets.
• To evaluate the performance of the YOLOv4-Tiny 3L architecture in comparison to other models for this specific application.

Main Methods:

• Utilized a video dataset for training and testing object detection models.
• Employed hyperparameter tuning, frozen layers, and data augmentation (photometric and geometric) on the YOLOv4 architecture.
• Compared the developed YOLOv4-Tiny 3L model against SSD-MobileNetV2 FPN and EfficientDet-D0.

Main Results:

• YOLOv4-Tiny 3L demonstrated superior performance for real-time detection.
• Achieved 90.56% mAP for single-class and 70.21% mAP for multi-class detection.
• Exhibited significantly faster detection speeds compared to YOLOv4-CSPDarknet53, SSD-MobileNetV2 FPN, and EfficientDet-D0.

Conclusions:

• YOLOv4-Tiny 3L is highly suitable for real-time oil palm detection tasks.
• The video-based approach and YOLOv4-Tiny 3L architecture enhance harvesting efficiency.
• This research advances automated systems in precision agriculture.