Abstract:

---

Rice (Oryza sativa L.) is a staple food for nearly half of the world’s population, and its productivity largely depends on the availability of essential nutrients such as nitrogen (N), phosphorus (P), and potassium (K), with deficiencies in these elements leading to reduced yield and food security risks. Traditional manual methods of diagnosing nutrient deficiencies are often inaccurate, time-consuming, and labor intensive, highlighting the need for automated, reliable solutions. This study proposes a deep learning based system for the automated identification of NPK deficiencies in rice leaves at early growth stages using a dataset of 1,155 images collected from Kaggle, where preprocessing involved resizing, normalization, augmentation, and class balancing with SMOTE. A baseline Convolutional Neural Network (CNN) was developed and compared with an enhanced CNN integrated with a Spatial Attention Mechanism (SAM), with the baseline achieving 96.5% accuracy and a 94.7% F1-score, while the CNN+SAM model significantly outperformed it, reaching 98.8% accuracy and a 98.2% F1-score. By enabling the model to focus on critical leaf regions, the attention mechanism improved feature discrimination and interpretability, demonstrating CNN+SAM as a robust and explainable solution for early detection of nutrient deficiencies and a valuable tool for advancing precision agriculture and sustainable rice production.

Keyward(s): EfficientNet, Machine Learning Classifiers, Nitrogen (N), Phosphorus (P), and Potassium (K), Natural Language Processing, Nutrient Deficiency, Spatial Attention

Download