Improved Monitoring of Wildlife Invasion through Data Augmentation by Extract-Append of a Segmented Entity

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Crop damage from wild animals is a growing problem in South Korea.
• Existing methods like fences and deterrents have limitations.
• Deep learning object detection offers potential but requires extensive, specific datasets.

Purpose of the Study:

• To address the challenge of limited datasets for wild animal detection models.
• To propose a novel data augmentation technique for improving model training.
• To enhance the performance of object detection models for protecting crops.

Main Methods:

• Developed an Extract-Append data augmentation method.
• Utilized semantic segmentation to extract specific wild animal objects (water deer, wild boar).
• Appended extracted objects to diverse background images to create an augmented dataset.

Main Results:

• Trained an object detector using the Extract-Append technique.
• Compared performance against existing data augmentation methods.
• Achieved a mean Average Precision (mAP) improvement of at least 2.2%.

Conclusions:

• The Extract-Append method effectively generates rich datasets for wild animal detection.
• This technique overcomes limitations posed by scarce, specific wild animal data.
• Significant improvements in detection performance demonstrate the method's practical value for farm protection.