Black gram disease classification using a novel deep convolutional neural network

Black gram, the king of pulses, also known as Urad in India, where it is cultivated from ancient times. India is the largest cultivator of black gram crop globally, and its production is declined year by year because of the diseases that occurred to the black gram plant leaves. The plant disease recognition and classification systems that rely on Convolutional Neural Networks (CNNs) have demonstrated encouraging results under limited and controlled circumstances. Such models are often constrained by a laeconomic losses to theck of consistency, making them less reliable. When detecting diseases with complicated background images taken from the cultivation field conditions, the models' accuracy would degrade drastically. To overcome this issue, firstly leaf region is segmented from all the images in the dataset using DeepLabv3 + layers with MobileNetV2 as a feature extractor. And then, the dataset was enhanced and expanded to 15000 images with the help of rotation, mirror symmetry, illumination correction, random shifting, and noise injection augmentation techniques. As a final step, proposed a Deep Convolutional Neural Network (DCNN) model for the identification and classification of black gram plant leaf diseases, taking into consideration of a large number of parameters, size, and depth of the available state-of-the-art CNNs. The proposed DCNN model was trained and tested on the segmented leaf regions from the images in Black gram Plant Leaf Disease (BPLD) Dataset, which was created from real cultivated fields. The 5-fold cross-validation technique was employed to check the proposed model’s efficiency for detecting the diseases in all the scenarios. The performance evaluation and the investigation outcomes evident that the proposed DCNN model surpasses the state-of-the-art CNN algorithms with 99.54% accuracy, 98.80% F1 score, 98.78% precision and 98.82% recall for black gram plant leaf diseases classification and can provide a pragmatic solution for real-world applications.