A method to solve the multi-frame overlapping error of UAV object recognition

Abstract

A method for solving the multi-frame overlapping error of UAV object recognition, the method comprising the following steps: Step 1, obtaining aerial image data of ground objects; Step 2, expanding the training data set; Step 3, training data preprocessing; Step 4, improve the YOLOV3 model loss function and output of detection results; Step 5, improve the YOLOV3 ground object detection model training; Step 6, online application of the ground object detection model. In order to avoid the problem of overlapping identification of illegal buildings, the present invention improves the loss function of the YOLOV3 model and the output of detection results, so that the model is more suitable for the identification of illegal buildings on the ground, and reduces the problem of motion blur in images taken by drones, and enhances the system. Adaptability and robustness are conducive to improving the control accuracy of the system.

Description

technical field [0001] The invention relates to the field of unmanned aerial vehicle detection, in particular to a method for solving the multi-frame overlapping error of unmanned aerial vehicle object recognition. Background technique [0002] Urban illegal buildings refer to buildings that violate the relevant laws and regulations of urban planning and land management, have not obtained the planning permission of construction land, or have changed the planning permission regulations of construction projects without authorization. Illegal buildings not only disrupt the order of urban planning and construction, illegally occupy public resources, affect the image of the city, and pose great safety hazards, but also damage the credibility of the government and affect the future development of the city. Therefore, how to discover and deal with illegal buildings in a timely, accurate and efficient manner has become an urgent problem for urban management departments. [0003] At...

AI Technical Summary

Problems solved by technology

Deep learning generally detects the target through the anchor frame. The anchor frame is generally pre-designed with parameters such as height and width, and then the network detects the target through the anchor frame. However, when multiple anchor frames detect the same target, the phenomenon of repeated target detection is prone to occur.

Images