Sustainable food production increasingly relies on intelligent automation technologies that optimize resource usage while maintaining crop health. This study presents a deep learning-driven computer vision and web based monitoring system designed for automated nutrient management in soilless (hydroponic) agriculture.
The proposed system integrates environmental sensors and a Raspberry Pi-based control unit to continuously monitor plant growth parameters and manage nutrient solutions through real-time feedback. A novel strawberry leaf image dataset was constructed to detect visual symptoms of nutrient deficiencies such as chlorosis and tip burn.
Three convolutional neural network (CNN) architectures such as ResNet-18, MobileNetV2, and DenseNet-121 were trained and evaluated to classify these deficiencies. Among them, MobileNetV2 achieved the highest performance with an overall test accuracy of 0.9967, demonstrating superior generalization across classes.
In addition, Grad-CAM visualizations confirmed that the model accurately localized key symptom regions on the leaves, enhancing interpretability and trustworthiness. A React-based web interface was developed to enable remote visualization of system data, sensor readings, and nutrient levels, allowing real-time monitoring and manual intervention when necessary.
The proposed system effectively integrates artificial intelligence, computer vision, and IoT-based monitoring to provide an efficient, automated, and sustainable solution for modern hydroponic farming.