Transfer learning (TL) enables knowledge learned in one domain to improve performance in another related domain; however, its application to industrial and medical anomaly detection (AD) is often limited by domain discrepancies and scarce labeled data. This research addresses these challenges by bridging Aircraft Fuselage Defect Detection (AFDD) and Chest X-ray (CXR) Abnormality Identification through cross-domain TL, enabling effective feature generalization between industrial and medical imaging tasks. Publicly available aircraft inspection images and CXR datasets are used to ensure diversity and representative anomalies. Data pre-processing incorporates adaptive histogram equalization (AHE) for contrast enhancement and median filtering (MF) to reduce noise, followed by image normalization to standardize input dimensions. For robust feature representation, Gray-Level Co-occurrence Matrix (GLCM) and Histogram of Oriented Gradients (HOG) are employed to capture complementary structural and texture information. The proposed Dynamic Grey Wolf Optimizer-driven Deep Convolutional Transform Network (DGWO-DCTN) integrates preprocessing, feature extraction, and AD into a unified framework. A DC Neural Network (DCNN) extracts hierarchical spatial features, while a Transformer module models long-range dependencies and global contextual relationships. To improve convergence and generalization, the DGWO adaptively tunes network hyperparameters and weights. TL is realized by fine-tuning the model on CXR data using pretrained knowledge from fuselage defect detection. Experimental evaluation implemented in Python demonstrates strong performance, achieving 94% precision, 93.75% recall, 96.80% accuracy, and a 93.80% F1-score. These results confirm that combining TL with hybrid deep architectures provides an effective and computationally efficient solution for cross-domain AD.
Keywords: Computer Vision, abnormality identification, cross-domain, industrial inspection, medical imaging
