Deep Convolutional Neural Networks

Over the last decade, deep learning frameworks have revolutionized the field of computer vision, delivering state-of-the-art performance across various applications. The advent of deep convolutional neural networks (CNNs) has placed deep learning at the forefront of several fields including Biomedical, Healthcare, Information and Communications Technology (ICT), Digital Technology (Digitech), and Industrial Automation. Deep learning is a data-driven approach which includes feature extraction network and task-specific module for downstream task. These tasks may include regression, classification, image segmentation, and object detection. Deep learning models involve two primary processes: Training and Testing phase. During the training phase, models learn from labeled or annotated data to make predictions. Training involves adjusting the parameters and biases of models to ensure their predictions closely match actual values, known as the ground truth. During the learning or training process, a quantified value, known as the objective loss score, is computed to assess the deviation from expected results. This score guides the framework in adjusting its weights to improve accuracy. Following training, the testing or inference phase evaluates the model’s performance on previously unseen and unlabeled data, determining its real-world applicability. This chapter provides an introduction to deep learning-based networks, with a concise explanation of common deep learning architectures used for image feature extraction, their key components, and a discussion of some recently introduced and widely utilized object detection frameworks.