Intelligent detection system of diseases in hydroponic crops using artificial vision

Abstract

This work proposes the development of an intelligent system for early disease detection in hydroponic lettuce crops using computer vision and artificial intelligence (AI). A model based on the MobileNetV2 architecture was implemented, it was trained using transfer learning and executed on a Raspberry Pi 4 to ensure portability and low resource consumption. The training stage was      with a set of 2,180 images classified into four classes (healthy, mildew, powdery mildew, and bacterial spot), using data augmentation and preprocessing techniques tailored to the model. An accuracy of 96.34% was achieved in the validation set and 93.64% in the test set, demonstrating high generalization capabilities. The model was integrated into a prototype that captures images online, performs local inferences, and provides immediate visual physical feedback about the detected disease. This research demonstrates the technical feasibility of applying AI to precision agriculture for hydroponic environments, reducing manual intervention and improving production efficiency, with the potential to scale to smart agricultural solutions.