Research On Dense Object Detection Algorithm Based On Convolutional Neural Networks

With the development of Internet and computer technology,the quantity of image data is increasing rapidly.Meanwhile,vision is the main source of human sensory information.Computer vision aims to simulate human vision system for computing,perceiving and cognizing.All these facts make computer vision become an important research field.Object detection is a fundamental visual recognition problem in computer vision and has been widely studied in the past decades.Visual object detection aims to find objects of certain target classes with precise localization in a given image and assign each object instance a corresponding class label.The emergence of convolutional neural networks has greatly improved the performance of object detection algorithm.However,with the continuous enrichment of application scenarios,object detection algorithm is required to be more accurate.The factors of multi-scale and dense target in object detection have been impacted on the performance seriously.In this context,this paper mainly researches two kinds of the dense object detection algorithm,which includes:Underwater object detection based on the class-weighted YOLO net is proposed.Underwater objects detection exists many issues,such as image blurring,various object scales and object densely.In this paper,we propose a class-weighted YOLO net for underwater object detection,in which a class-weighted loss is designed to balance sample of difficulty so as to acquire better result.Moreover,a dimension adaptive clustering of object box is introduced to promote the detection performance.The experimental results show that the proposed method outperforms the traditional YOLO net in the task of dense object detection which every image nearly contained 20 objects.Multi-orientation remote sensing object detection based on attentional mechanism is proposed.The rapid development of remote sensing techniques has significantly increased the quantity and quality of remote sensing images.in this context,significant efforts have been made in the past few years to develop a variety of methods for object detection in optical remote sensing images.Object detection in remote sensing images is an active yet challenging task in computer vision because of the random orientation of objects brings from bird view perspective,the highly complex backgrounds,and the variant appearances of objects.Especially when detecting densely packed objects in remote sensing,images.This paper based on Faster R-CNN,attention module is introduced to solve the small dense object.Moreover,we use the method of image pyramid to reduce the impact of multi-scale objects.In the experiments,the comparison between this method and other methods shows that the proposed method effectively improves the performance of the orient remote sensing object detection algorithm.