Edge AI-Enabled Chicken Health Detection Based on Enhanced FCOS-Lite and Knowledge Distillation

TL;DR

A real-time and compact edge-AI enabled detector designed to identify chickens and their healthy statuses using frames captured by a lightweight and intelligent camera equipped with an edge-AI enabled CMOS sensor that meets practical demands for automated poultry health monitoring using lightweight intelligent cameras with low power consumption and minimal bandwidth costs is presented.

Abstract

The utilization of AIoT technology has become a crucial trend in modern poultry management, offering the potential to optimize farming operations and reduce human workloads. This paper presents a real-time and compact edge-AI enabled detector designed to identify chickens and their healthy statuses using frames captured by a lightweight and intelligent camera equipped with an edge-AI enabled CMOS sensor. To ensure efficient deployment of the proposed compact detector within the memory-constrained edge-AI enabled CMOS sensor, we employ a FCOS-Lite detector leveraging MobileNet as the backbone. To mitigate the issue of reduced accuracy in compact edge-AI detectors without incurring additional inference costs, we propose a gradient weighting loss function as classification loss and introduce CIOU loss function as localization loss. Additionally, we propose a knowledge distillation scheme to transfer valuable information from a large teacher detector to the proposed FCOS-Lite detector, thereby enhancing its performance while preserving a compact model size. Experimental results demonstrate the proposed edge-AI enabled detector achieves commendable performance metrics, including a mean average precision (mAP) of 95.1$\%$ and an F1-score of 94.2$\%$, etc. Notably, the proposed detector can be efficiently deployed and operates at a speed exceeding 20 FPS on the edge-AI enabled CMOS sensor, achieved through int8 quantization. That meets practical demands for automated poultry health monitoring using lightweight intelligent cameras with low power consumption and minimal bandwidth costs.

Figures (11)

• Figure 1: Schematic of the edge-AI enabled detector. During the training phase, the compact FCOS-Lite detector, acting as the student model, is improved through knowledge distillation and tailored detection loss functions, then following compression for inference, the refined student model is deployed on the edge-AI enabled CMOS sensor.
• Figure 2: Example of the whole system featuring light-weighted intelligent cameras and our proposed detector: (a) Overall system placement in a real-world AIoT scenario, (b) Intelligent camera (left) and its internal edge-AI enabled CMOS sensor (right), and (c) Example of a visual result outputted by the proposed detector.
• Figure 3: An example of the training dataset: (a) and (b) show high-quality images of healthy and sick chickens, respectively, (c) and (d) display healthy and sick chickens captured from real scenarios, respectively, with annotation labels included in (d).
• Figure 4: FCOS-Lite network structure.
• Figure 5: An example of weights based on the gradient norms for classification loss, with thresholds $\mu$ are set to 0.4, 0.6, 0.8. respectively.