Molecular validation integrated reliable deep learning approach for early blight severity detection in tomato crops under protected cultivation environments

• Novel molecular validation approach ensures reliable data for deep learning models. • Evaluated five pre-trained CNN models and a custom 18-layer CNN model for early blight severity detection in tomatoes. • DenseNet121 achieved 99.61% accuracy and an F1 score of 0.993, proving its effectiveness in real-time detection. • Transfer learning with pre-trained models outperformed custom models. • Grad-CAM visualizations enhanced interpretability by highlighting key regions for disease severity classification. High disease infestation under protected cultivation environments causes substantial yield losses and poses challenges for reliable identification due to resembling symptoms among different diseases. A molecularly validated deep learning framework was developed for precise detection and severity classification of early blight in tomatoes caused by Alternaria alternata . Pathogenic identity of fungus was confirmed through molecular validation, ensuring dataset reliability. An image database comprising 6,531 tomato leaf images across four severity levels (healthy, low, medium, and high) was developed under diverse lighting conditions, which were augmented to 12,131 images. Transfer learning was applied to five pre-trained CNN models—MobileNetV2, DenseNet121, InceptionV3, VGG19, and NASNetMobile—alongside a custom 18-layer CNN model. Among these, DenseNet121 outperformed others, achieving classification accuracy of 99.61%. Grad-CAM visualization further validated model’s decision-making by highlighting critical image regions for severity classification. Inference time of just 0.19s makes DenseNet121 most suitable for real-time control of early blight. Molecular validation of diseased samples confirmed a 99.44% identity with Alternaria alternata isolates, reinforcing the reliability of dataset and model predictions. This reliable disease detection system based on molecular validation can be implemented for targeted spray application in protected cultivation. This approach can help in resource-efficient and sustainable management strategies under multiple diseases.