A non-invasive, image-based monitoring system for vermicomposting developed as an academic research project at Iloilo Science and Technology University (ISAT-U). Given a captured image of a vermicompost substrate, the system predicts the substrate weight in real time using a CNN regression model — eliminating the need for manual inspection or physical disturbance of the worm habitat.

The pipeline extracts additional training frames from short video captures using OpenCV, feeds them through an EfficientNet-B0 backend model or a MobileNetV3-Small TFLite model for on-device inference, and surfaces predictions through a unified Flutter app targeting both mobile and web from a single codebase.

• InstitutionIloilo Science and Technology University (ISAT-U)
• PlatformMobile & Web (Flutter)
• StackFlutter, FastAPI, TensorFlow / Keras, MobileNetV3-Small (TFLite), EfficientNet-B0, Python, OpenCV
• ModelCNN Regression — single numeric output (substrate weight)
• DeploymentTFLite (on-device mobile), SavedModel / ONNX (backend)