Lightweight Reverse Tampon Target Detection Algorithm Based on YOLOv8

In the medicating process of cotton swabs, accurately identifying the orientation of the swabs is crucial to ensuring uniform absorption of the medication. Traditional object detection algorithms are often constrained by their dependency on large-scale datasets and substantial computational resource consumption. To overcome these limitations, this paper introduces a lightweight YOLOv8 framework-based algorithm specifically designed for detecting reversed cotton swabs. The algorithm initially incorporates the concept of partial convolution, which effectively reduces the parameter count. It then employs a Cross-Scale Feature Fusion Module (CCFM) to optimize the network architecture of YOLOv8, aiming to enhance object detection performance while maintaining high computational efficiency. Furthermore, the algorithm integrates a focal modulation module to boost model performance and utilizes the latest boundary loss function, Minimum Point Distance Intersection over Union (MPDIoU), to improve boundary regression outcomes. Tested on a custom-built dataset, the algorithm achieves an average detection accuracy of 96.42%, increases the frame rate by 23.1 milliseconds, and maintains a model size of just 3.4MB. These results provide effective technical support for the precise and rapid identification and handling of medical cotton swabs, offering substantial practical value.