Cataract is the clouding of the eye lens and is a major cause of blindness across the globe. Detecting cataracts early and prompt treatment can prevent blindness. To reduce dependence on experts to examine the eye fundus image, computer-assisted technologies are useful for early diagnosis. In the past, many cataract detection models have been described. Some of these models employed typical machine learning techniques, but the performance of these models is inefficient to make them useful. The image classification field has seen much improvement with the help of deep learning. One of the key drawbacks of these cutting-edge models is that they require a significant amount of computational resources and time for training. This problem is solved via transfer learning, which allows pre-trained models to be used for feature extraction. We present an ensemble technique for cataract diagnosis using the eye fundus image based on VGG-19, ResNet101V2, and InceptionV3. Soft voting was used to determine the final classification. The ensemble model gave the F-1 Score of 95.90 on the test dataset. According to our results, this ensemble network is more accurate than any single network.