<![CDATA[Surgical instrument recognition is a critical component in robot-assisted surgery for precisely identifying the types of instruments being utilized. This study investigates the application of contour-based geometric descriptors to recognize surgical instruments without requiring a model training process (training-free). The proposed approach extracts instrument contours and calculates three geometric descriptors: Aspect Ratio, Solidity, and Contour Ratio. The descriptor values of the input objects are compared against an offline reference descriptor matrix using the Euclidean distance measure. The main contributions of this study lie in proposing a lightweight and training-free surgical instrument recognition scheme utilizing a combination of three contour-based geometric descriptors, as well as providing a transparent and traceable (explainable) classification decision mechanism based on Euclidean distance similarity comparison across classes. Evaluation was performed using a primary laboratory dataset acquired with a Logitech Brio 4K camera (1080p, 60 fps) at a distance of 30 cm, an illumination of 300 lux, and a plain green cloth background. The testing dataset comprised 10 images containing 30 controlled instrument instances (10 instances each for Tissue Forceps, Dressing Forceps, and Mayo-Hegar Needle Holder). On this specific testing dataset, the proposed method achieved a recognition accuracy of 100% with an average processing time of 59.31 ms per image. Although obtaining perfect accuracy under controlled laboratory conditions, these results are interpreted strictly within the limits of the experimental environment and do not yet reflect generalization to complex real-world surgical images.]]>
