OBJECT DETECTION USING SSD AND EFFICIENT NET B7 AS BASE NETWORK

Abstract

This thesis addresses the complex challenge of image classification and object detection using advanced deep learning techniques. The research focuses on utilizing EfficientNet B7 for image classification and implementing the Single Shot MultiBox Detector (SSD) for object detection. The study initially evaluates the performance of four neural networks—VGG-16, ResNet-50, AlexNet, and EfficientNet-B7—on an animal dataset containing 31 distinct classes. For the image classification task, each model was trained and tested to determine its accuracy in recognizing various animal species within the dataset. Among the evaluated models, EfficientNet-B7 emerged as the superior performer, achieving the highest accuracy in both training and testing phases. This outstanding performance underscores the model's capability to effectively handle complex classification tasks and its potential for broader applications in the field of computer vision. Building on the success of EfficientNet-B7 in image classification, the research proceeded to integrate this network as the foundational layer for the SSD framework in object detection tasks. The SSD leverages the feature extraction capabilities of EfficientNet-B7 to detect and localize objects within images. The combination of EfficientNet-B7's robust feature extraction and SSD's efficient detection mechanism resulted in an object detection accuracy of approximately 78%. The findings of this thesis highlight the efficacy of EfficientNet-B7 in both image classification and object detection domains. By demonstrating superior performance in classification tasks and achieving notable accuracy in object detection, this research contributes valuable insights into the application of deep learning models for complex computer vision challenges. The study provides a comprehensive evaluation of neural network models and offers a compelling case for the adoption of EfficientNet-B7 and SSD in practical image analysis and object detection scenarios. This work lays the groundwork for future research and development in the field, emphasizing the importance ofselecting and optimizing deep learning models for specific tasks to achieve the best possible outcomes.