Grape diseases and pest infestations threaten the economic viability of grape production, necessitating early detection and intervention. Leveraging advancements in machine learning and computer vision, researchers are developing automated systems that accurately identify and classify grape diseases, contributing to improved disease management strategies. This study proposes an automated framework for classifying and detecting grape leaf diseases, integrating an enhanced metaheuristic optimization algorithm with deep learning techniques. To address the class imbalance present in the Grape dataset, a Conditional Generative Adversarial Network (CGAN) is employed as a data augmentation technique, generating synthetic images to balance the representation of each class. Two pre-trained convolutional neural network (CNN) models, AlexNet and ResNet18, are then utilized to extract deep features from the augmented images. A fusion method aggregates the extracted feature vectors, which are subsequently optimized using an improved metaheuristic optimization algorithm for feature selection (FS). Metaheuristic algorithms, known for their dynamic search behavior and global search capabilities, offer promising solutions for FS. This study introduces the Enhanced Chameleon Swarm Optimizer (ECSA) method, a novel variant of the metaheuristic Chameleon Swarm Algorithm (CSA), to address the FS problem. The ECSA, with its use of chaotic maps during the exploration phase and integration of Levy flight distribution into the exploitation phase, represents a significant advancement in metaheuristic optimization. The final set of selected features is then classified using the K-Nearest Neighbors (KNN) algorithm for grape leaf disease identification. The performance of the proposed framework is assessed on a real-world dataset of grape diseases, employing multiple evaluation criteria. The proposed framework demonstrates superior performance, achieving a peak accuracy of 97.76% on the grape disease dataset.