Wastewater treatment plants offer an important pathway for microplastics (MPs) to enter natural aquatic systems. Traditional MP analysis often involves manual sorting of MPs, which can be time-consuming and labor-intensive and is susceptible to human errors, particularly when handling large sample sizes. Herein, five convolutional neural networks—EfficientNet_b7, Inception_v3, Resnet_v2_50, Resnet_v2_101, and Mobilenet_v3—were employed using transfer learning (TL) methods to categorize MPs extracted from municipal wastewater treatment plants into four morphologies: fibers, films, fragments, and pellets. A Samsung Galaxy S22 Android smartphone was used to capture pictures of the MPs using a stereomicroscope, which were then used to train the neural networks. First, TL methods were employed using an 80:20 split of training and validation datasets and testing sets, and the effectiveness of the methods was evaluated via five-fold cross validation. Second, the effect of the sizes of the training and validation sets on the performances of different neural networks was estimated through augmentation of these sets to 300% and their combination with the original dataset, yielding 5104 images. The overall performances of the considered neural networks were evaluated in terms of accuracy, precision, recall, F1 score, and training time. The proposed approach achieved accuracies of 92%–96% and 94%–98% in classifying the original and augmented datasets, respectively, based on MP morphologies. Notably, EfficientNet_b7, Inception_v3, and Mobilenet_v3 achieved the highest classification accuracy of 98% when trained on the augmented dataset. Mobilenet_v3 achieved a competitive accuracy on our datasets while being considerably smaller and more computationally efficient, leading to its selection for deployment in Android mobile and web applications. Overall, Mobilenet_v3 can promptly classify MPs in real time, offering an efficient, effective solution for MP image classification on a smartphone.