Detecting plant leaves disease is essential for maintaining healthy agricultural production and averting serious damage from various diseases. Conventional diagnostic techniques, however, can be challenging and time-consuming. The proposed work initially analyzes the use of the convolutional neural network, namely VGG16, VGG19, ResNet50, InceptionV3, MobileNetV2, and EfficientNetB0 for disease detection and its severity classification, due to their strong feature extraction and classification capabilities. The work then fine-tunes these models and proposes ensemble-learning-based pear leaves disease classification with a combination of two and three top-performing models for an increase in efficiency. Even though there are a number of two models and three model ensemble learning, the performance of the combination of two models ensembling MobileNetV2 + InceptionV3 emerged as the best overall performer with an accuracy of 90.20%, precision of 90.96%, recall of 90.20%, and F1 score of 89.75% respectively, and performs well with disease severity classification as well.