A Novel Deep Convolutional Neural Network for Efficient Classification of Lettuce Diseases

Diseases pose a significant threat to crop yields in vertical hydroponic farms worldwide. The lack of specialized knowledge makes plant disease detection a challenging and complex task. Leveraging DL models for disease identification from leaf images provides a promising solution. However, these models require substantial training data and computational resources. In this research work a novel Conv-7 deep convolutional neural network (DCNN) has been proposed for classifying lettuce leaves into different categories with enhanced accuracy. The Conv-7 DCNN model consists of seven convolutional and pooling layers, followed by three fully connected layers. This architecture enables the model to learn rich feature representations from the images, resulting in more accurate disease classification. The model is capable of categorizing leaves into three classes: bacterial, fungal, and healthy. To prevent overfitting during training, dropout rate of 0.2 has been considered, batch normalization, L2 regularization with a value of 0.0001, and image augmentation techniques such as shifting, shearing, scaling, zooming, and rotation. A publicly available Kaggle dataset of hydroponically grown lettuce leaves has been utilized to train the CNN model. The proposed Conv-7 DCNN achieves superior performance compared to several DL architectures such as VGG-16, VGG-19, ResNet-50V2, MobileNet, DenseNet-121, and InceptionV3. This model attains magnificent metrics such as accuracy of 97%, precision of 97%, f1-score of 97%, recall of 97%, MCC (Matthews Correlation Coefficient) of 96%, FPR (False Positive Rate) of 1.36 %, FNR ((False Negative Rate) of 3.14%, NPV (Negative Predictive Value) of 99%, FDR (False Discovery Rate) of 2.8%, specificity of 99%, and an AUC (area under the curve) of 1. Moreover, the integration of the proposed model with IoT devices can enable real-time applications in agricultural fields.