DISubNet: Depthwise Separable Inception Subnetwork for Pig Treatment Classification Using Thermal Data
The classification of thermal images has recently drawn attention in various applications, including autonomous vehicles, security surveillance systems, face recognition, etc. Since the visible spectrum has many limitations, such as object shadows, similar colour backgrounds, and varying lighting conditions, thermal imaging can overcome these issues. Thermal images are increasingly used to detect diseases and distress in poultry, swine,
and dairy animal husbandry. In this project, we developed a depthwise separable inception subnetwork (DISubNet), a lightweight model for classifying four pig treatments. Based on the modified model architecture, we propose two DISub- Net versions: DISubNetV1 and DISubNetV2. Our proposed models were compared to other deep learning models commonly employed for image classification. The thermal dataset captured by a forward-looking infrared (FLIR) camera was used to train these models. The experimental results demonstrate that the proposed models for thermal images of various pig treatments outperform other models. In addition, both proposed models achieve approximately 99.96%– 99.98% classification accuracy with fewer parameters.
In addition to multi output classification regarding the various farming practices and animal handling in the pig industry, the developed model aims to develop a framework for decision support system for predictive analytics for determining variations in the pig behaviour with respect to change in external perturbations such as the play time, feeding interval time and resting time. This crucial information eventually helps the animal scientists in recommending best practices for enhancing animal welfare.
