Healthy-unhealthy animal detection using semi-supervised generative adversarial network

Area of Science:

• Veterinary Medicine
• Computer Science
• Machine Learning

Background:

• Animal illness poses risks to human health through disease transmission.
• Early human history shows recognition of animals as disease carriers.
• Detecting unhealthy animals is crucial for public health and animal welfare.

Purpose of the Study:

• To develop an automated system for detecting healthy versus unhealthy animals.
• To improve upon existing methods for animal health assessment using computer vision.
• To address the challenge of limited datasets in animal health image analysis.

Main Methods:

• Utilized a deep learning approach for animal health detection.
• Employed image augmentation techniques (flipping, scaling, orientation) to expand the dataset.
• Implemented a fuzzy-based brightness correction method.
• Applied a semi-supervised generative adversarial network (SGAN) for classification.

Main Results:

• Achieved 91.4% accuracy in detecting healthy and unhealthy animals.
• The method demonstrated efficacy on an augmented COCO dataset.
• The system accurately identifies animal health status from images.

Conclusions:

• A novel two-fold system for animal health detection was successfully developed.
• The proposed system significantly advances the field of automated animal health assessment.
• The approach is adaptable for various computer vision applications requiring animal health analysis.