Wiener filter based deep convolutional network approach for classification of satellite images

摘要

Semantic segmentation is a fundamental task in computer vision and image scenery detection. Many applications, such as urban planning, change detection, and environmental monitoring require accurate segmentation. Hence, most segmentation tasks are performed by humans. Currently, with the growth of deep convolutional neural network (DCNN), there are many works aimed to find the best network architecture fitting for this task. In this work, the GoogLeNet classifier is used to perform better segmentation as well as a classification for satellite images. The Wiener filter is used here for image denoising. Data Augmentation is performed to extract high information about the input picture. The output of the above steps helps in classification i.e. it identifies the scenery of the input image with four labels. The result shows that the GoogLeNet based image classification has reduced error rate and it also increases the accuracy of output. Additionally, the efficiency of the Wiener filters also described clearly in the result.