Satellite Image Classification based on Adaptive Skip Connection - Convolutional Neural Network

Satellite image classification is required in many fields such as weather forecasting, urban planning, monitoring natural resources, and mapping. Existing satellite image classification have the limitation of overfitting problem that degrades the performance of the model. This research involves in applying Adaptive Skip Connection (ASC) technique in Convolutional Neural Network (CNN) model for satellite image classification. The Min-Max normalization technique is applied to normalize the images of the dataset. The ASC technique skip the connection if the probability of the overfitting is high and connection is not skipped if more similar images are present. The ASC-CNN model is tested on SAT-4 dataset and compared with the existing techniques. The ASC-CNN model is compared with the deep learning techniques and the ASC-CNN model has higher efficiency than existing methods. The ASC-CNN model has 95.4 % accuracy and 13 layer CNN model has 93.1 % accuracy in satellite image classification.