Research on Residual Welding Rod Detection Technology Based on Deep Learning

In the domain of welding, conventional methods for quantifying residual welding rods predominantly depend on manual counting, which is not only labor-intensive but also susceptible to inaccuracies. To address this issue, this study proposes an instance segmentation model based on the improved Mask Region-based Convolutional Neural Network model, referred to as Point Mask R-CNN. By integrating an attention mechanism and the PointRend module, the model synergizes channel and spatial attention within the convolutional neural network, thereby augmenting its perceptual capabilities. Furthermore, it refines boundary delineation and detail segmentation through dynamic point sampling and localized refinement processes. Experimental findings demonstrate that this model not only marginally elevates the detection efficacy of the basic model but also markedly enhances segmentation performance, facilitating a more efficient and precise identification of residual welding rods.