CLASSIFICATION OF TRANSMISSION TOWERS IN SATELLITE IMAGES BASED ON TRANSFERRED CONVOLUTIONAL NEURAL NETWORK

In this paper the utilization of transferred deep learning techniques for transmission tower image classification is investigated using Alexnet and Googlenet. The database upon fine tuning of the overall images is acquired from Google Earth and internet totaling to 1300 images specifically 650 are images of transmission tower whilst another 650 images are non-transmission tower. Here, 600 images are chosen at random as the training dataset for fine-tuned deep learning neural networks and the balance namely 50 images are used for validation. The same dataset is used as input to both networks. In addition, both networks own similar training setting too. Results attained showed that Alexnet and Googlenet are capable to perform this classification task with perfect classification by Alexnet specifically 100% while Google obtained 99% accuracy rate. As for computational performance, Googlenet performed faster as compared to Alexnet with 787 minutes by Googlenet versus 853 minutes by Alexnet.