Nutrient deficiency detection in hydroponic Turkish cucumber using ensemble technique

The cultivation of hydroponic plants has gained increasing importance due to factors such as limited space requirements, improved control over environmental conditions, and the elimination of pesticide use. However, even in controlled or semi-controlled environments, nutrient deficiencies can still occur, adversely impacting plant growth and nutritional value. This paper proposes an image-based deep learning approach using ensemble feature extraction combined with a long short-term memory (LSTM) classifier to detect nutrient deficiencies in hydroponic plants. Among the ensemble combinations tested, the model integrating VGG16, ResNet18, and ResNet34 achieves the best results, with 90.67% accuracy on the test set. The deep learning-based explainable AI Grad-CAM++ method is used to promote interpretability and trust in AI-driven decisions. Multiple optimizers are also evaluated, with AdaBelief yielding the most stable and accurate training outcomes. The proposed model provides a lightweight, sensor-free solution that aids farmers in the early detection and management of nutrient deficiencies in hydroponic systems.