In this project, we proposed a simple application to predict the cooking time of beans. The underlying model is based on MobileNetV2. The results obtained show that the model identifies the variety and predicts the corresponding cooking time with an accuracy of 86%.
Key Technical Aspects
Architecture: The use of MobileNetV2 as a backbone allows for a lightweight and efficient model, ideal for integration into mobile devices with limited computational resources.
Dual-Task Performance: The system successfully bridges the gap between classification (identifying the specific bean variety) and regression (estimating the continuous value of the cooking time).
Performance: Achieving an 86% accuracy rate indicates a strong correlation between the visual morphological features of the beans and their physical hardness/thermal resistance.
Impact of the Solution
By providing an instantaneous prediction, this application directly addresses energy management challenges. It empowers users to optimize their use of cooking fuels—such as gas, charcoal, or wood—by providing a clear expectation of the preparation duration, thereby enhancing efficiency in both domestic and semi-industrial culinary contexts.