Enhanced Weld Defect Categorization via Nature-Inspired Optimization-Driven Neural Networks

In many different sectors, the safety and dependability of welded structures are crucially dependent on the accurate identification of weld defects. To address this, we present a novel method for categorizing weld defects named as Pelican Optimization-based convolutional Bi-Long Short-Term Memory Networks (POCBi-LSTM). With this integrated strategy, weld defect categorization is intended to be more accurate and effective. To train and assess a suggested classification algorithm, a collection of 220 samples with 5various types of weld faults is utilized. The study's technique comprised denoising with a Modified Median Filter (MMF) for noise reduction and binary histogram equalization (BHE) for image enhancement. The Scale Invariant Feature Transform (SIFT) was used to extract features. The outcomes demonstrate that the proposed strategy performs greater than conventional methods and delivers superior recognition accuracy, precision, recall, and F1 scores of 92.9%, 94.5%, 89.5%, and 92.8%, respectively . Experiment findings demonstrate the effectiveness and advantage of the recommended approach in classifying weld faults. Through the use of an integrated strategy, it is hoped to increase the dependability and effectiveness of weld defect categorization by handling local pattern recognition and long-term interdependence efficiently.