Building surface disease detection method and system based on convolutional neural network

Abstract

The invention provides a building surface disease detection method and system based on a deep learning network model. The method comprises the following steps: acquiring a building surface image as a data set; inputting the data set into a deep learning network model for learning, wherein the deep learning network model detects and fuses the multi-scale feature map of the feature extraction network in the learning process; carrying out primary iteration training on the fused feature map in the deep learning network model, carrying out secondary training according to a cosine annealing learning rate in a set range after the primary training is completed, storing parameters of the model iterated each time in the secondary training, and solving medians of all models to obtain a new model; and then identifying building surface diseases based on the trained deep learning network model. According to the method, relatively small disease features can be identified, the model AP accuracy and the target positioning and classification accuracy are greatly improved, and the identification of the disease features is more accurate.

Description

technical field [0001] The invention relates to the field of deep learning, in particular to a method and system for detecting building surface defects based on a convolutional neural network. Background technique [0002] The reliability, safety and integrity of buildings are critical to the well-being of society, so detection of damage to building surfaces is extremely important. Taking bridges as an example, detecting bridge surface diseases can effectively prevent bridge wear, promote bridge maintenance and increase bridge service life. [0003] However, in the current technology of identifying and monitoring non-destructive diseases of bridges, manual visual inspection is the main method, which results in low efficiency, time-consuming and labor-intensive, and subjective evaluation. In this context, computer vision-based detection technology has been applied and developed, which uses wall-climbing robots or drones to obtain bridge images, and uses machine learning algo...

AI Technical Summary

Problems solved by technology

The current Anchor-Based target detection algorithms are generally divided into two categories. One is the region-based two-stage target detection algorithm Faster-RCNN [1] and Mask-RCNN [2]. It integrates defect feature extraction, region proposal network, bounding box regression, etc., resulting in slow model operation and poor real-time performance

For example, Cha et al. [3] used Faster R-CNN to detect and quantify five types of surface damage in reinforced concrete bridges. Although good results were obtained, the detection speed was not ideal

The other is the single-stage algorithm SSD[4] and YOLO series, which use the regression idea to directly mark the position and category of the target image. This type of algorithm makes up for the shortcomings of the two-stage target detection algorithm based on the area, and has a large Improvement, but a slight decrease in accuracy, especially the SSD algorithm cannot make full use of the shallow high-resolution feature map, making the recognition accuracy not ideal

Images