The field of music emotion recognition (MER) is quickly gaining traction as it attempts to identify and categorize the emotions expressed via music. In this study, we propose a novel hybrid deep learning method called MER_Net, which combines an LSTM model for improved MER accuracy with a customized convolutional neural network (CNN). Leveraging two benchmark MER datasets namely, multi-modal MIREX dataset and 4-quadrant emotion dataset, we achieved notable recognition accuracies of 98.46% and 96.25%, respectively. We provide comprehensive results analysis including comparison with state-of-the-art MER methods. Additionally, we discuss limitations, future improvements, and emphasize the motivational significance of bridging gaps in MER research.