Enhancing the adaptability of YOLOv8 for waste detection

With the acceleration of urbanization and population growth, the quantity of waste is rapidly increasing, imposing immense pressure on the environment, resources, and public health. Garbage classification and detection is a significant topic in the field of computer vision. However, due to the diverse types of garbage, varying shapes, complex backgrounds, and lighting changes, existing object detection methods often perform poorly in addressing these issues. YOLOv8, as an advanced object detection algorithm, has shown outstanding performance in multiple tasks due to its efficient detection capabilities and accuracy. Nevertheless, its direct application in garbage detection is suboptimal. Therefore, this study proposes a garbage detection method based on YOLOv8, named EA-YOLO. Firstly, this study introduces an innovative Adaptive Inner-IOU mechanism, which can adjust the proportion factor of Inner-IOU adaptively according to the needs of different detection tasks, enabling the model to better focus on hard examples. Secondly, due to the complexity and diversity of garbage detection, a new attention mechanism, ISRA, is introduced in this paper. This mechanism allows the model to dynamically adjust the attention to different regions based on the characteristics of the targets and environmental changes, thus adapting to garbage object detection tasks in various scenarios. Finally, by introducing deformable convolution, this paper proposes a new feature fusion module, FlexiC2f, to enhance the flexibility of feature fusion. This module can dynamically adjust the deformation of convolution kernels according to the shape and position of the input features, thereby achieving more precise and efficient feature fusion. Experiments were conducted on the WaRP dataset. The experimental results show that on the WaRP dataset, the proposed method improves the [email protected] (%) by 6.1% and the [email protected]:0.95 (%) by 5.1% compared to the existing optimal methods, proving the effectiveness of the proposed method.