Contour Decomposition and Line Segment Feature Recognition of 2D Images Based on Bi-LSTM

Using a high-energy beam to cut patterns according to the contours in images is a typical application. The basis for generating G-code for numerical control systems is the efficient decomposition of contour lines into continuous line segments and the determination of their attributes. However, current traditional methods and machine learning algorithms are complex and have low accuracy. To resolve the shortcomings and limitations of current methods, this work proposes a contour segmentation and feature recognition method that combines traditional image processing methods with Bidirectional Long Short-Term Memory (Bi-LSTM) based machine learning models. Experimental comparisons between the proposed model and traditional methods on authentic 2D images indicate that the model presented in this study exhibits stronger robustness. The proposed method achieves a detection accuracy exceeding 84%, significantly outperforming the compared end-to-end neural network approaches. Additionally, it demonstrates a faster recognition speed. The applicability of this method for contour extraction tasks in industrial scenarios has been verified.