Hybrid deep learning framework for high-precision, real-time monitoring of cauliflower crop health using spectral data

Pest infestations and diseases such as black rot caused by Xanthomonas campestris pv. Campestris pose significant threats to cauliflower cultivation, often leading to severe yield losses. To address this challenge, a hybrid deep learning framework was developed for real-time crop health monitoring under field conditions. Cauliflower (Brassica oleracea var. botrytis) was cultivated under black rot–conducive conditions and inoculated using the leaf cut-and-dip method at IARI, New Delhi. Spectral data, including RGB and NDVI values, were collected every three days after symptom onset using a TCS3200 color sensor and a Greenseeker device. The dataset, consisting of 2,110 spectral data counts each of healthy and diseased cauliflower plants, was pre-processed using mean substitution for missing values, normalized with MinMaxScaler, and reshaped for model compatibility. A hybrid deep learning framework was employed, combining Convolutional Neural Networks (CNNs) for spatial feature extraction, Long Short-Term Memory (LSTM) models with attention mechanisms for temporal analysis, and a Random Forest (RF) classifier for final classification. The CNN achieved an accuracy of 94.54%, while the LSTM with attention outperformed it with 97.14% accuracy. The integrated CNN-LSTM-RF model demonstrated outstanding performance, achieving 98.81% accuracy, with precision, recall, and F1-scores of 0.99 for both healthy and diseased crops. This study underscores the transformative potential of the proposed hybrid framework in enabling precise, real-time crop health monitoring. By reducing chemical inputs, improving disease management, and enhancing productivity, the framework supports sustainable agricultural practices. Its scalability across diverse agro-climatic conditions positions it as a promising tool for global food security.