Attention deficit hyperactivity disorder (ADHD) is a neurodevelopmental disorder that often begins at childhood and can persist into adulthood. Children with this disability often cannot sit still in a place and have to bear academic losses due to inattention and behavioral spontaneity, which inhibit them in executing demanding or repetitive tasks. Time frequency image (TFI) of electroencephalogram (EEG) was used in a deep learning model to detect ADHD, and the effects of theta (4–8 Hz) and upper- beta and lower- gamma (13–40 Hz) waves in an ADHD patient were compared. Results were further supported with gray level co-occurrence matrix (GLCM)-based machine learning approach and statistical analysis. A TFI feature extraction network for ADHD detection based on convolutional neural network (CNN) was designed and trained. The model achieved 95.36% overall test accuracy on ADHD detection with test accuracy of 99.75% and 91.25% at the 4–8 Hz and 13–40 Hz bands, respectively. Textural features, including the GLCM features, were extracted from the TFIs and were learned by using a K-nearest neighbor (KNN) classifier. An overall classification accuracy of 93.16% was obtained by the KNN with 96.5% and 90.00% test accuracy at the 4–8 Hz and 13–40 Hz bands, respectively. The statistical analysis with ANOVA showed significant results of <0.001 for all the 13 features studied. The TFI-based CNN model and GLCM-based KNN classifier supported higher ADHD-related information at the theta band compared with the upper- beta and lower- gamma bands.