Computer vision serves as a foundational pillar in the domain of digital image processing, permeating diverse applications such as visual object detection, intelligent surveillance, pedestrian detection, autonomous driving, automatic picking, and industrial inspection. It exploits the computational capabilities of computers to automate functions that were traditionally conducted manually, ushering in substantial implications for conserving human resources. Within computer vision, visual object detection emerges as a pivotal element, extensively employed in numerous applications including, but not limited to, face recognition, gait analysis, segmentation, and pedestrian identification. This thesis delves into deep learning-based visual object detection approaches, which are fundamentally segmented into three categories: Two-stage, one-stage, and transformer-based object detection methods. We have incorporated these methodologies to facilitate the recognition and categorization of fruits, utilizing the transformer-based model to attain a remarkable accuracy rate 99% in fruit classification. Furthermore, our model manifests the capability to execute accurate recognition within a mere 0.12 seconds. The insights derived from this exploration hold potential to augment the efficiency and applicability of computer vision in varied contexts, furthering the advancement of this multifaceted field.