Henrique Fernandes

Breast cancer detection is a global health priority. While traditional methods have limitations, infrared thermography offers a promising, non-invasive alternative by detecting subtle thermal changes that can indicate tumors. This paper assessed five pre-trained Convolutional Neural Networks (CNNs) for breast cancer detection using DMR-IR thermal images, employing a 5-fold cross-validation. Among the tested models, ResNet50 achieved the best overall performance, with the highest average accuracy (92.79%), precision (95.00%), specificity (98.67%), sensitivity (72.00%), and F1-score (79.43%). The model was trained using raw thermal images from three anatomical views (frontal, lateral 90°, and lateral 45°), totaling five images per patient, an approach still uncommon in the literature. These results highlight ResNet50's strong potential for reliable and clinically applicable breast cancer detection using thermography.

Interpreting chest X-ray images can be challenging, leading to an increasing reliance on computational methods by physicians to improve diagnostic accuracy and disease monitoring. Among these methods, machine learning models—particularly those in computer vision, such as Convolutional Neural Networks (CNNs)—have shown strong performance in image classification tasks. Recently, Kolmogorov-Arnold Networks (KANs) have emerged as an alternative framework that redefines how connections and activations are computed within neural architectures. This study aims to develop a classification system using a dataset of 5,000 chest X-ray images categorized as Normal, Pneumonia, or Tuberculosis. Two KAN-based algorithms, Efficient KAN (EKAN) and Fast KAN (FKAN), were implemented and compared with a CNN baseline. Experimental results show that the FKAN model achieved performance comparable to CNN with approximately 96% accuracy in test cases, but with a reduced computational cost of approximately 96.18%. The results suggest that the FKAN and EKAN algorithms are a practical and efficient approach for medical image classification.

This study evaluates the effectiveness of thermographic signal reconstruction using polynomial approximations for preprocessing thermographic images obtained through Pulsed Thermography in the segmentation of defects in CFRP laminates. Neural networks were trained with both processed and unprocessed images and evaluated using IoU and F1-Score metrics. The preprocessing significantly increased defects' signal-to-noise ratio and detection accuracy, achieving an IoU of 92% and an F1-Score of 98%. The technique proved effective in enhancing defect segmentation and definition, particularly in scenarios with low signal-to-noise ratio.

Detecting and quantifying barely visible impact damage (BVID) in carbon fiber-reinforced polymer (CFRP) materials is a key challenge in maintaining the safety and reliability of composite structures. This study presents the application of low-power vibrothermography to identify and quantify such defects. Using a long-wave infrared (LWIR) camera, thermal data were captured from the CFRP specimens that inhibit BVID. How image processing, specifically principal component analysis (PCA) and sparse principal component analysis (SPCA), can enhance thermal contrast and improve the accuracy of defect size is also explored. By combining low-energy excitation with advanced data analysis, this research aims to develop a more accessible and reliable approach to non-destructive testing (NDT) for composite materials.

The coffee roasting process is a critical factor in determining the final quality of the beverage, influencing its flavour, aroma, and acidity. Traditionally, roast‐level classification has relied on manual inspection, which is time‐consuming, subjective, and prone to inconsistencies. However, advancements in machine learning (ML) and computer vision, particularly convolutional neural networks (CNNs), have shown great promise in automating and improving the accuracy of this process. This study evaluates multiple ML models for coffee roast level classification, including a CNN with Xception as a feature extractor, alongside AdaBoost, random forest (RF), and support vector machine (SVM). The models were trained and tested on a public dataset of 1,600 high‐quality images, balanced across four roast levels: green, light, medium, and dark, to ensure robust performance. Experimental results demonstrate that all models achieved 100 % accuracy and F‐1 scores, confirming their effectiveness in accurately distinguishing roast levels. Furthermore, the proposed approach was compared with previous studies, showing strong performance in roast classification. Image augmentation techniques were applied to improve generalizability in real‐world applications. This research presents a reliable, scalable, and fully automated solution for roast‐level classification, significantly contributing to quality control in the coffee industry. Practical Applications: This research offers a reliable and automated way to classify coffee bean roast levels using image analysis and ML. It can help coffee producers and roasters improve quality control by providing faster, more consistent, and objective assessments of roast levels, ultimately ensuring a better product for consumers.

This study explores the application of low-power vibrothermography (LVT) for detecting barely visible impact damage (BVID) in carbon fibre-reinforced polymer (CFRP) laminates. Composite specimens with varying impact energies (2.5–20 J) were excited using a single piezoelectric transducer with a nominal centre frequency of 28 kHz, operated at a fixed excitation frequency of 28 kHz. Thermal data were captured using an infrared camera. To enhance defect visibility and suppress background noise, the raw thermal sequences were processed using principal component analysis (PCA) and robust principal component analysis (RPCA). In LVT, RPCA and PCA provided comparable signal-to-noise ratios (SNR), with no consistent advantage for either method across all cases. In contrast, for pulsed thermography (PT) data, RPCA consistently resulted in higher SNR values, except for one sample. The LVT results were further validated by comparison with PT and phased array ultrasonic testing (PAUT) data to confirm the location and shape of detected damage. These findings demonstrate that LVT, when combined with PCA or RPCA, offers a reliable method for identifying BVID and can support safer, more efficient structural health monitoring of composite materials.

The aviation industry currently faces several challenges in inspecting and diagnosing aircraft structures. Current aircraft inspection methods still need to be fully automated, making early detection and precise sizing of defects difficult. Researchers have expressed concerns about current aircraft inspections, citing safety, maintenance costs, and reliability issues. The next generation of aircraft inspection leverages semi-autonomous and fully autonomous systems integrating robotic technologies with advanced Non-Destructive Testing (NDT) methods. Active Thermography (AT) is an example of an NDT method widely used for non-invasive aircraft inspection to detect surface and near-surface defects, such as delamination, debonding, corrosion, impact damage, and cracks. It is suitable for both metallic and non-metallic materials and does not require a coupling agent or direct contact with the test piece, minimising contamination. Visual inspection using an RGB camera is another well-known non-contact NDT method capable of detecting surface defects. A newer option for NDT in aircraft maintenance is 3D scanning, which uses laser or LiDAR (Light Detection and Ranging) technologies. This method offers several advantages, including non-contact operation, high accuracy, and rapid data collection. It is effective across various materials and shapes, enabling the creation of detailed 3D models. An alternative approach to laser and LiDAR technologies is photogrammetry. Photogrammetry is cost-effective in comparison with laser and LiDAR technologies. It can acquire high-resolution texture and colour information, which is especially important in the field of maintenance inspection. In this proposed approach, an automated vision-based damage evaluation system will be developed capable of detecting and characterising defects in metallic and composite aircraft specimens by analysing 3D data acquired using an RGB camera and a IRT camera through photogrammetry. Such a combined approach is expected to improve defect detection accuracy, reduce aircraft downtime and operational costs, improve reliability and safety and minimise human error.