Understanding human emotions through EEG signal analysis plays a vital role in improving brain-computer interfaces and advancing mental health applications. EEG signals provide comprehensive insights into emotional states such as arousal, representing activity or calmness, and valence, indicating pleasantness or unpleasantness. However, interpreting these signals remains a challenge due to their complexity. To address this, we propose a Deep Convolutional Regression Network (DCRN) that predicts valence and arousal from EEG signals, turning them into spatial heatmap representations. These images serve as input for the DCRN, enabling continuous predictions of emotional states. By establishing thresholds, these predictions allow the classification of emotions into distinct categories: active, pleasant, inactive, and unpleasant. This approach offers significant advantages in emotion recognition accuracy, benefiting areas such as mental health and human-computer interaction. The proposed DCRN was tested on the widely recognized DEAP dataset for emotion analysis and demonstrated remarkable performance, with a 96% accuracy in predicting valence and arousal. The model effectively transforms EEG signals into spatial data, enabling precise emotion classification. In conclusion, the DCRN has shown excellent capability in regressing both arousal and valence values, achieving low final losses of 0.0245 for arousal and 0.0781 for valence. The close alignment of training and validation losses confirms the robustness and reliability of the model. With its high classification accuracy and ability to generalize complex emotional patterns, the DCRN model is well-suited for real-time emotion recognition, enhancing the potential of brain-computer interfaces and contributing to advancements in affective computing.