Guava is a crop commodity in West Java with total production in 2021 reaching 692,488 quintals. This production decreased by -12.82% compared to 2020 which amounted to 794,345 quintals. The research uses deep learning technology with the Convolutional Neural Network (CNN) algorithm and MobileNetV2 architecture to classify digital images of guava leaves and fruits that have been labeled or called supervised learning. The development method used in this research is Cross Industry Standard Process for Data Mining (CRISP-DM). Based on the results of this study, the guava leaf model has excellent evaluation results, training accuracy of 99.6%, validation accuracy of 100%, training loss of 3.2%, and validation loss of 3.1%. The confusion matrix of this model has 100% accuracy from 63 validation data. Meanwhile, the guava fruit model requires a dropout of 0.2 and L2 kernel regularizers of 0.01 to reduce overfitting. This model has a training accuracy of 98.8%, validation accuracy of 91.6%, training loss of 19.1%, and validation loss of 38.6%. The confusion matrix results show that the accuracy of this model reaches 91.6% of 84 validation data. Then the model was successfully implemented into a mobile-based application using the Kotlin programming language.