<![CDATA[The growing demand for efficient environmental monitoring has led to the development of computer vision based trash detection models for real world urban deployment. These systems must recognize diverse waste types under challenging conditions, such as variations in appearance, scale, and complex backgrounds, which can significantly reduce detection accuracy. Moreover practical applications require lightweight models capable of maintaining real-time performance on CPU devices. This study enhances the YOLOv12n architecture for trash detection by integrating a Reinforced Local Multi-Scale Channel Attention Module (RL-MSCAM). The proposed module improves feature representation by highlighting informative channels and reducing irrelevant background noise, leading to better extraction of discriminative spatial features of waste objects. The lightweight design ensures these improvements are achieved without significant computational overhead, making the model suitable for real-time applications. Experimental results show that the proposed model achieves strong detection performance, with 60.6 % mAP@50 and 50.9% mAP@50:95. Despite its lightweight design, the model maintains practical efficiency, reaching an inference speed of 16.12 frames per second (FPS). These findings indicate that integrating RL-MSCAM into the YOLOv12n framework improves trash detection performance while preserving computational efficiency, making it suitable for real-time deployment on CPU devices.]]>
