In recent years, the number of pneumoconiosis patients has been increasing worldwide, placing a significant burden on physicians for diagnosis. As a result, there is growing interest in supporting diagnosis through Computer-Aided Diagnosis (CAD), with convolutional neural networks (CNNs) in particular demonstrating high diagnostic accuracy. In this study, we propose a method to further improve the accuracy of pneumoconiosis image classification using CNNs by inputting images processed with wavelet transformation. Prior to inputting wavelet-transformed images, we also select an appropriate optimization algorithm and learning rate. By comparing classification accuracy under different combinations of optimization algorithms and learning rates, we identify the most suitable combination for the task. Furthermore, while previous studies have proposed the use of images decomposed using the Haar wavelet transform, we compare classification accuracies using three additional wavelet function families. Finally, based on the results from the two experiments, we use Gradient-weighted Class Activation Mapping (Grad-CAM) to visualize the basis of image classification for the models that achieved the highest accuracy in each experiment.