Quantum Deep Convolutional Generative Adversarial Networks Based Denoising with Modified Convolutional Neural Network-Linear Discriminant Analysis for Early Disease Detection in Grape Leaves
Abstract
Grape farming entails the growth of grapevines to be used in diverse ways, but the leaves of the grape are likely affected by diseases that may easily spread in an ideal environment. However the existing techniques are prone to failure in the coexistence of multiple diseases in which the presence of more than one disease on a given leaf and absence of time analysis implies that there is no better representation of how the disease progresses with time and deeper layers of some methods can lose valuable spatial and texture data, which can result in the incomplete representation of the disease characteristics. To overcome from these constrains, proposed a modified convolutional neural network-linear discriminant analysis (MCNN-LDA) for detecting the disease of grape leaves earlier. Initially, these process starts with the collection of grape leaf disease dataset and then preprocessed through quantum deep convolutional generative adversarial networks (QDCGAN) technique, which are used for removing the unwanted noises from the image and texture- contrast enhancement fusion network (TCEFNet) which are used to enhance the quality of images. Then the preprocessed image are fed into the segmentation phase, where a Convolutional block attention module-fusion enhanced transformer-Unet (CBAM-FET-Unet) is used to isolate a region of interest with an added CBAM module to focus on the most important features. The segmented image will be classified by MCNN-LDA, in which the traditional ReLU activation function will be exchanged with the Advanced Exponential Linear Unit (AELU) to enable better feature extraction and less information loss. This system classifies grape leaf diseases into four groups such as Healthy, Black Rot, ESCA, and Leaf Blight, which offer an effective and precise system of early detection of the diseases. The proposed technique outperforms with better classification results, containing the accuracy value of 97.07%, Error value of 2.9% and precision value of 94.8% respectively. The proposed MCNN-LDA model is an effective and precise approach of detecting grape leaf disease in the real-time, and improve the health of grapevines in real agro-ecology context.
