ايوب اميني

With rapid advances in satellite and sensor technologies, there is a dramatic increase in access to remote sensing images. However, exploring these images involves a huge amount of human intervention that is tedious, time consuming, and inefficient. There is always a need to develop automated detection algorithms to help imaging professionals access complete information from images and locations to find objects of interest in a more accurate and efficient manner. To fully understand these images, it is important for image analysts to know if there is an interest in it. As a result, object recognition has been a major problem that plays a role in many applications, such as geographic mapping, urban planning, vehicle tracking and environmental monitoring. In the dissertation, we address the problems of object recognition in aerial and satellite imagery applications and provide an object recognition algorithm for high-aerial imagery with the location that the algorithm is designed in terms of number and volume of computations relative to the rest of the algorithm Is Multiple applications have good productivity and provide a program simultaneously and with good accuracy. Due to the importance of speed, accuracy and precision in Object recognition algorithms which is the main purpose of this dissertation, in this application we consider EfficientDet object recognition algorithm as the base network and by improving this network, object recognition algorithm, which is able to Design to predict objects with rotational angles (Oriented) in satellite images. To eva‎luate and eva‎luate the network in this dissertation from the famous DOTA dataset that is presented in the following years for object detection. Is used and first eva‎luate the optimized algorithm by performing the operation as preprocessing the images of this data set and preparing the appropriate images and tags with them to the appropriate network input site.

تشخيص اشياء , تشخيص هواپيما , شبكه هاي كانولوشني عميق , تشخيص اشياء چرخشي , تصاوير ماهواره اي

object detection , airplane detection , Rotated object detection , deep convolutional networks , satellite images

Ayub Amini

Mohammad Soleimani