An algorithm for optimizing psychological regulation strategies for college students based on image recognition and...

Intelligent and adaptable psychological management systems are needed due to the growing academic pressure and emotional problems experienced by college students. Traditional methods, relying on self-reported questionnaires and manual counselling, are often subjective and lack real-time flexibility. The research aims to develop an algorithm for optimizing psychological regulation strategies using image recognition and reinforcement learning, incorporating a Weighted Butterfly-Twin Delayed Deep Deterministic Policy Gradient (WB-TwinD3PG) Model. Emotional and psychological data are obtained from a publicly available dataset containing around 10,000 records with more than 20 features representing diverse psychological states. Two pre-processing methods are applied: face detection and alignment to separate and normalize facial regions, and illumination normalization to minimize lighting variations. For feature extraction, Local Binary Patterns (LBP) are employed to extract discriminative texture-based features indicative of emotional states, such as stress, anxiety, neutrality, and relaxation. The reinforcement learning module engages WB-TwinD3PG to dynamically optimize psychological regulation strategies. This advanced algorithm develops an exploration-exploitation balance, accelerates convergence, and increases policy stability by integrating WB-inspired optimization with TwinD3PG. The agent selects personalized interventions, including relaxation exercises, cognitive prompts, and motivational feedback, based on emotional state feedback. The combination of WB-TwinD3PG enables adaptive, robust, and personalized mental health support for college students. According to experimental data, the suggested model outperformed traditional techniques in terms of recall (0.921), F1-score (0.927), precision (0.934), and accuracy (0.913). Overall, this research provides a practical framework that combines image-based emotion analysis with advanced reinforcement learning for psychological well-being management.