Traffic infrastructure construction target monitoring system and method based on target anchor frame optimization strategy

Abstract

The invention discloses a traffic infrastructure construction target monitoring system and method based on a target anchor frame optimization strategy, and the method comprises the following steps: 1, carrying out the remote sensing image information collection through a remote sensing image processing technology, and feeding back the information to an internal image preprocessing module; 2, carrying out related preprocessing operations such as image enhancement and normalization on the acquired image by an image preprocessing module; 3, automatically inputting a preprocessing result to a convolution feature extraction module, extracting corresponding multi-level multi-scale convolution features, and carrying out ROI region positioning and feature extraction through the provided candidate box tuning scheme; and 4, carrying out target frame fine processing by using the extracted candidate frame score tuning and candidate frame position tuning. According to the method, the accurate and reliable optimal candidate box selection strategy is constructed by using the information of a plurality of reliable candidate boxes in the image neighborhood, the positioning and classification precision of the capital construction main body target in the high-resolution remote sensing image is improved, the candidate box tuning strategy can effectively extract the detection box with the real target, and missing detection of part of small-scale targets in the detection process is also relieved.

Description

technical field [0001] The invention relates to the technical field of traffic infrastructure construction monitoring, in particular to a traffic infrastructure construction target monitoring system and method based on a target anchor frame optimization strategy. Background technique [0002] At present, the two-stage target detection method based on anchor frame extraction and classification is still the mainstream solution for precise target positioning and classification, such as R-CNN [1], Fast R-CNN, Faster R-CNN. This type of method mainly places anchor boxes of different scales and aspect ratios densely on the feature map candidate area, and then classifies and regresses these anchor boxes according to the acquired feature representation. Usually, after the target anchor box is matched and thresholded, a target object will have multiple candidate anchor boxes for rough registration, so it is necessary to use non-maximum suppression processing to filter the candidate r...

AI Technical Summary

Problems solved by technology

The use of the enhanced feature map in this patent is the key point but also its defect. This is mainly because the enhanced algorithm is a preset solution and cannot be adaptively adjusted according to the feature distribution and target diversity of the image in the subsequent detection process.

Therefore, the positioning accuracy of the detection frame depends on the algorithm preset scheme, which cannot be applied to diverse building objects

Images