Enhancing Precision Agriculture with a CNN-RF Hybrid Model: A Case Study on Sunflower Disease Diagnosis

Disease classification in sunflower crops is an important step in mitigating the yield losses incurred because of high disease spread and improving agricultural productivity. In this paper we present a novel hybrid model that combines Convolutional Neural Networks (CNNs) for feature extraction and Random Forest (RF) for classification so as to leverage the complementary strengths of these techniques. The model was evaluated on a sunflower disease dataset containing 10,000 images across five classes: Downy Mildew, Rust, Leaf Blight, and Alternaria Leaf Spot are healthy. The hybrid CNN-RF model yielded an accuracy of 96.2 %, beating standalone CNN (93.5<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>) and RF (85.1%). Robustness and a tradeoff between false positives and false negatives are also demonstrated by the model, evidenced by the precision., recall, and F1 scores of 96.5%., 96.0%, and 96.2%, respectively. The performance in discriminative capability was excellent, with an AUC of 0.95 yielded from the receiver operating characteristic (ROC) curve. A confusion matrix analysis showed that over 95% of cases were accurately classified, while the highest misclassification rate of only 3% was observed between Rust and Alternaria Leaf Spot. The hybrid model was thoroughly trained with minimal overfitting, as shown by the losses of 0.12 and 0.18., and ended with 98.2% and 96.2% final training and validation accuracies. Although with an inference time of 25ms per image, it would also be suitable for real-time agricultural uses. The results show that the CNN-RF hybrid model is a suitable, scalable., and efficient solution for automated sunflower disease detection with the potential to revolutionize precision agriculture.