This research work proposes a novel technique to classify fruits based on photographs by introducing the convolutional neural network (CNN) model VGG19 combined with a blurred layer to increase the accuracy of identifying fruits from a user-provided dataset consisting of five types of fruit: apples, bananas, mangoes, pineapples, and oranges. Specifically, we added a blur layer to preprocess the input images, which allows for more robust and precise identification due to the use of VGG19 pretrained model and further tuning of the model. We then applied transfer learning by fully training only the last few layers of VGG19, leaving the remaining layers frozen, to allow for tuning the VGG19 architecture for classifying fruits in our specific task. Augmentation techniques such as random shearing, zooming, and flipping horizontally ae applied to our training dataset to promote further generalization and robustness of the model. In the evaluation of the model, we achieved over 90 percent accuracy for predicting fruits in the unseen test dataset, supporting the use of deep learning methods to solve real-world object recognition problems. Recognition, particularly with an emphasis on the need for better modeling methods when classifying the fruit in scarce environments and identifying fruit in real time. In addition to recognizing the value of transfer learning, these results highlight the importance of custom modeling, which can be used for high-quality classification tasks through performance-enhancing techniques.