Research on intelligent detection method of construction quality based on UAV image recognition and convolution neural network

Construction quality is the key factor to ensure the safety and durability of the project. There are many disadvantages in traditional detection methods, which are difficult to meet the demand of efficient, accurate and objective detection in the current digital transformation of the construction industry. This study focuses on the integration of UAV image recognition technology and Convolutional Neural Network (CNN), and innovatively constructs a closed-loop system for intelligent detection of construction quality. At the data level, a comprehensive UAV image dataset covering 12 types of defects and five major engineering scenarios was carefully constructed. In algorithm design, the MF-ASPPNet (Multi-Feature ASPP Network) architecture was proposed, utilizing ResNet-18 as the backbone and incorporating an improved feature pyramid network (FPN) for multi-scale feature fusion. The atrous spatial pyramid pooling (ASPP) module was introduced to expand the receptive field and capture multi-scale information, while the CBAM attention mechanism was integrated to accurately focus on key defect regions and suppress background noise, significantly improving detection accuracy and localization performance. In addition, in order to adapt to the deployment of UAV edge devices, the lightweight strategies such as deep separable convolution instead of standard convolution and channel pruning are implemented, which significantly reduces the calculation and parameters of the model and accelerates the reasoning process. In the experimental stage, a diversified data set containing 15,680 images was selected. After rigorous training, verification and testing, MF-ASPPNet surpassed the mainstream detection model in key indicators such as [email protected]:0.95, and the crack detection accuracy reached 78.4%, the reasoning speed reached 37 FPS, and the model parameters and calculation amount were controlled at a low level, showing excellent detection performance and efficient deployment ability, which strongly promoted the construction quality detection direction.