Edge-AI: Optimization of Fishpond Health Classification System, Using Image Recognition
Year: 2026
Type: conference-paper
Abstract: This research introduced a lightweight Edge-AI fishpond health classification system. That enables real-time evaluation of pond conditions directly on a mobile device. Using image recognition, the system processes, uploaded or captured photos through the smartphone camera. The application classifies pond status as healthy, unhealthy and unknown entirely offline. Transfer learning was used to build the model, as a pretrained deep neural network. This was fine-tuned on 1,347 fishpond images to efficiently learn pond-health classification. This ensures fast, low-power inference suitable for rural environments with limited connectivity. The system architecture integrated a mobile-based image sensing, local preprocessing and edge-AI inference. It delivered instantaneous prediction without additional hardware or sensors. Experimental results show reliable performance across varied pond images with average prediction confidence score of 99.6%. This demonstrated the viability of mobile-based edge intelligence for aquaculture monitoring. While the current model focuses on binary classification, the work establishes a foundation for future expansions that can involve multi-class diagnosis, multimodal sensing, explainable-AI integration and automated pond management.
Language: English
Subfield: Water Science and Technology
Field: Environmental Science
Domain: Physical Sciences
SDG: Zero hunger
Open Access status: hybrid
OpenAlex ID: W7215965124