India's agriculture permits the world food chain by producing various crops and boosting the country's economy. Diseases pose a significant challenge to agricultural production. It causes crop disruption, lowers output, and makes it extremely hard for farmers to compensate for planting damage. Early disease detection and rapid action are essential to preventing productivity loss. Currently, several methods for analyzing illness characteristics and figuring out the stage of progression use Machine Learning (ML) for image processing. However, because disease features vary, it is challenging to determine the regional segments. Unbalanced traits can complicate the detection of diseases. To resolve this problem, initially, we collected the plant image dataset from Kaggle. We applied preprocessing steps, including Gaussian and Wiener filters, to normalize plant leaves. Furthermore, plant leaf features can be selected using the Canny Region Extraction (CRE) technique for non-edge and smoothing. Moreover, the Multilevel Threshold Segmentation (MRDS) method can identify pixel groups and classify the optimal values. Finally, the proposed ResNet50 Optimal Convolutional Neural Network (ROCNN) method can categories the results to obtain binary plant classification. As a result, accuracy for plant leaf diseases can be obtained using high false rates, imprecise recognition, high precision, F-measure and low recall efficiency.