Improved YOLOv8 Algorithm for Indoor Object Detection

Abstract To address the challenges of missed detections and false positives in indoor object detection tasks caused by target clustering and occlusion, this study proposes an advanced indoor object detection algorithm, SRSE-YOLOv8. A novel indoor object detection algorithm, SRSE-YOLOv8, is proposed to address missed and false detections caused by object clustering and occlusion in indoor scenes. First, an additional smaller detection head is introduced to enhance the capture of small objects and improve feature extraction capabilities. Then, RFAConv is incorporated to achieve input-adaptive kernel adjustment through multi-scale receptive field learning and spatial-channel attention coordination, thereby enhancing the model’s ability to handle complex images. Next, the Slimneck structure is applied in the neck network to reduce model complexity while maintaining accuracy. Finally, the EMA attention mechanism is introduced to filter out background noise interference and enhance feature responses in target regions. The proposed algorithm achieves an average precision of 94.9% on a self-constructed indoor waste dataset, representing a 3.5% improvement over the original YOLOv8n model. This approach effectively enhances indoor waste detection accuracy while meeting real-time detection speed requirements.