Research On Small Target Detection Algorithm Of UAV Aerial Image Based On Improved SSD

In recent years,the target detection technology based on deep learning has developed rapidly and has been widely used in the fields of security,autonomous driving and intelligent monitoring.As a comprehensive technology that combines positioning and recognition,the existing target detection model can well complete the object detection task in ordinary scenes,but when faced with the task of aerial image scene detection with complex background environment and many small objects,there are still huge challenges,especially the accuracy of small object detection is often not ideal,and there are a lot of missed detections.algorithm was improved,and the SSDF-C(SSD with Feature Fusion And Channel Attention)algorithm was proposed.The main work is as follows:(1)For the problem that the object size in the aerial image is relatively small,the lack of appearance information leads to the problem of low detection accuracy of small objects,Through multi-scale feature fusion,SSD-FC makes the shallow feature layer with delicate object features contain deep feature information,and then predicts the object,which can effectively avoid the problem of insufficient information of small objects when predicting the feature layer,thereby improving small objects.The accuracy of object detection;in the face of complex aerial images,the information of small objects will be masked by the noise of other larger objects,and even integrated with the surrounding environment,making it difficult to distinguish.In response to this,SSD-F-C introduces a visual attention mechanism,which allows the model to focus more on small objects,reduces the interference of small objects in complex environments,strengthens the information of small objects,and reduces the missed detection rate of small objects.(2)In order to enable the model to obtain better deep-level features in the feature extraction stage,SSD-F-C adopts Res Net-50 as the basic feature extraction network.The feature extraction capability is enhanced and the model operation efficiency is also improved to a certain extent.(3)In order to test the effect of the proposed algorithm in the specific aerial image detection task,the aerial data set was self-made according to the actual task,and the application of SSD-FC was studied,and the Datangxia UAV intelligent inspection system was constructed.Real-time monitoring of abnormal objects.The algorithm in this paper is compared with some current mainstream algorithms on NWPU VHR-10 and PALSCAL VOC and data sets respectively.The experimental results show that the performance of the algorithm in this paper is better than some current mainstream algorithms when comparing multiple detection indicators..And it can also quickly and accurately locate and identify the target object on the self-made application aerial data set.