Lightweight Zero-Deep Curve Estimation and Optimised Loss Function Analysis for Night-Time Image Enhancement using Multi-Source IoT Data

The Internet of Things (IoT) is a network of devices that exchange information autonomously via the web, enabling real-time monitoring, automatic operations, and decision-making based on received data. The research is about the creation of the Lightweight Zero-DCE Enhancement with Optimised Loss Using Multi-Source IoT Data (LZONMI) framework to enhance the precision and consistency of nighttime cattle detection under low-light and complicated conditions. A multi-source sensing system that combines different data types, such as GNSS measurements, meteorological data, Unmanned Aerial Vehicles (UAVs), images, trolley inspections, and 3D spatial modelling. Among the main ideas are an improved low-light correction method with a lightweight Zero-DCE network and a modified YOLOv5 detection model with a CBAM attention mechanism for more efficient feature extraction. A loss function based on Optimal Transport Assignment (OTA) that helps to optimise the stability of the training and accuracy of the localisation. The evaluation metrics are loss of 12, accuracy of 0.10, F1-score of 0.97 at IoU 0.50 and 0.85 at IoU 0.75, accuracy of 68.14%, and precision & recall of 0.91 to 0.99 calculations.