FSENet: Applying Deep Learning to Transparent Lens Counting

The accurate identification and counting of transparent lenses is critical in automated production processes, as it directly impacts overall production efficiency. However, current neural network models exhibit limitations in feature extraction, robustness against light interference, and handling sparse features of transparent lenses, making them inadequate for practical applications. Moreover, the task of counting transparent lenses presents several challenges, including blurred edges, lack of distinct textures, and interference from optical reflections. To address these challenges, we propose FSENet. Experimental results demonstrate that our model surpasses existing approaches in the counting task, yielding significant improvements in the MAE and RMSE metrics, with values of 14.54 and 18.47, respectively. These results indicate that FSENet offers enhanced accuracy and stability, positioning it as a more reliable solution for transparent lens counting tasks.