Convolutional Neural Network and Deep Learning Approach for Image Detection and Identification

Abstract

There are many different varieties of clouds, each with a unique set of properties. As a result of this variability, it is difficult to discern these sorts of clouds. A database's objects must be categorized using data categorization in order to be organized into multiple categories. This study made use of the Cirrus Cumulus Stratus Nimbus (CCSN) dataset, which falls under the low cloud category and includes photos of Cumulus (182 images), and Cumulonimbus (242 photographs), and Stratus (242 images) (202 images). A fast R-CNN detector with feature extraction = Resnet50 was used to create a system for classifying cloud kinds. A significant amount of training time is saved by the quicker R-CNN due to its lack of a selective search algorithm. Training loss values for cloud images had an average of 0.9030 from the first epoch through the last one. Using the Faster R-CNN object detection method with the Resnet50 architecture, cloud photos were added and the accuracy was 94.12 and the average precision was 0.76. - Faster R-advantages CNN affect the architecture utilized and are marginally influenced by the algorithm choice, however CNN with Resnet50 is superior overall where these advantages are held.