Multi-volume variable scale bitmap data object classification algorithm architectural concrete color difference detection

Abstract

Architectural concrete color difference (ACCD) directly affects building aesthetics. At present, the AC apparent quality detection method is based on a subjective artificial observation and detection process. The large volume of AC building, however, makes it difficult for quality inspectors to sample on site. Therefore, the manuscript establishes the aerial vehicle-to-ground coordinate conversion model and quadrotor unmanned aerial vehicle (QUAV) Eulerian motion model wherein the real-time optimization of QUAV motion parameters through remote sensing technology improves data collection stability. The proposed ACCD detection model, which is based on the QUAV remote sensing technology, is established by the improved multi-volume bitmap data object classification algorithm (BDOCA). The detection accuracy of MobileNetV3, ResNet50, and VGG16 algorithms reach 86.33%, 92.10%, and 87.62%, respectively. The study acquires experimental images from different buildings and produces a generalized identification set. It then verifies the ACCD detection model generalized identification capability by the traversal type detection method. The MobileNetV3, ResNet50, and VGG16 algorithms all achieved a generalized identification accuracy of 99.6%. The study builds the ACCD model by red-green-blue (RGB) spatial conversion, Kalman filtering noise reduction, noise point open operation, connected domain close operation, and mask quantization segmentation operation. This method implements color difference block coding, color difference threshold pixel point scale calculation, and geometric parameter calculation.

Keywords

• architectural concrete (ac)
• hyperparameter improvement
• variable scale feature fusion
• multi-volume algorithm comparison
• bitmap data