PROJECT TITLE :

CDD-Net A Context-Driven Detection Network for Multiclass Object Detection

ABSTRACT:

In contrast to object detection in natural images, which typically achieves a high level of success, the process of detecting and localizing multiclass objects in remote sensing imagery presents its own unique challenges. These challenges include large-scale change, uncertain direction, and high density. When it comes to resolving these issues with remote sensing images, having information about the context of the objects is extremely helpful. To improve the precision of multiclass object detection in remote sensing images, we suggest using a context-driven detection network (CDD-Net) in this letter. In order to learn the local context of the region of interest and thus capture the local neighboring objects and features, a local context feature network, abbreviated LCFN, has been proposed as a solution. In the meantime, a hybrid attention pyramid network (HAPN) is being designed. This network is able to direct attention to features that are more valuable. The feature pyramid network is enhanced with the addition of a squeeze and excitation block (SEB) and three asymmetric convolution blocks (ACBs) courtesy of the HAPN (FPN). Experimenting with the CDD-Net proposed here on the DOTA-v1.5 data set reveals that the results it produces are quite encouraging overall.