ResNet Convolution Neural Network Based Hyperspectral Imagery Classification for Accurate Cancerous Region Detection

Classification of cancer image based on Region of Interest (ROI) is the central issues of hyperspectral application. Using the available spatial information on images, the classification is a difficult task. However, due to the advancement of image processing algorithm, the processing of mixed pixel image is a notable research topic. Most of the classification techniques use only dimension reduction and depends on reference method. In this research, we merged both spectral and spatial characteristics information's about classification of mixed pixel image was presented. First, we perform training in the standard cancerous image dataset. Then proposed a deep classification architecture framework based on the Convolution Neural Network (CNN) based ResNet architecture of cancer region detection. Then in testing we present the image for classification based on ROI. To verify the performance of designed ResNet based CNN network, we calculated the performance index like accuracy, training time and classification error for detecting region of interest calcification. Then, we compared the performance with other conventional classifiers for experimental verification to oral cancer region detection. From the obtained results, we identified that the designed ResNet based CNN network can accurately classify the oral cancer in a mixed pixel complex image.