Enhancing Object Detection and Classification with Thermal Imaging: A Deep Learning and Optimization-Based Framework

The proposed research presents a robust and efficient framework for object detection and classification, leveraging advanced deep learning and optimization techniques. Object detection and classification tasks face challenges such as low-quality input data, inefficient feature extraction, suboptimal feature selection, and limitations in real-time performance. Pre-processing is often hampered by noise, inconsistent image scaling, and poor contrast, while traditional feature extraction and selection methods may fail to highlight and retain critical features adequately. Additionally, segmentation and classification often lack the precision required to meet the demands of real-world applications. To address these challenges, the proposed methodology incorporates a comprehensive pre-processing pipeline, including greyscaling, cropping, resizing, noise removal, image normalization, and contrast enhancement, ensuring high-quality input data. Improved CornerNet is employed for precise and effective feature extraction, capturing salient features crucial for downstream tasks. The Binary Sand Cat Swarm Optimization Algorithm (BSCSOA) is utilized for feature selection, efficiently reducing dimensionality while retaining relevant features. For detection, the Improved YOLOv9 Network ensures high accuracy and real-time performance in object identification. The Stacked Convolutional Bidirectional Long Short-Term Memory (SCBi-LSTM) network is used for classification, leveraging both spatial and temporal dependencies to achieve superior predictive performance. Finally, the Enhanced Manta-Ray Foraging Optimization Algorithm (EMRFOA) is applied for segmentation, delivering precise delineation of object boundaries. This integrated approach overcomes the identified challenges by enhancing data quality, optimizing feature selection, and improving detection, segmentation, and classification processes. It demonstrates significant improvements in accuracy and efficiency, making it highly suitable for diverse applications requiring advanced object detection and classification capabilities.