This paper presents a strategy to autonomously explore unknown indoor environments, focusing on 3D mapping of the environment and performing grid level semantic labeling to identify all available objects. Unlike conventional exploration techniques that utilize geometric heuristics and information gain theory on an occupancy grid map, the work presented in this paper considers semantic information, such as the class of objects, in order to gear the exploration towards environmental segmentation and object labeling. The proposed approach utilizes deep learning to map 2D semantically segmented images into 3D semantic point clouds that encapsulate both occupancy and semantic annotations. A next-best-view exploration algorithm is employed to iteratively explore and label all the objects in the environment using a novel utility function that balances exploration and semantic object labeling. The proposed strategy was evaluated in a realistically simulated indoor environment, and results were benchmarked against other exploration strategies.