Weighted Hermitian Wavelet Multilayer Extreme Learning Machine for Building Detection Using Unmanned Aerial Vehicle Images

Robust building extraction is crucial for 3D building models in Unmanned Aerial Vehicle (UAV) images. Automatic building extraction is in extremely high demand due to the productivity gain. Conventional building detection methods were time-consuming, costly, and highly complex for human experts. Several DL techniques were developed for building detection. But the accuracy was not enhanced. To solve this issue, the Weighted Hermitian Wavelet Multilayer Extreme Learning Machine (WHWMELM) technique is proposed with multiple layers for building detection in UAV images. The aim of the proposed method is to accurately detect building objects with maximum accuracy and minimal time. First, the input layer obtains UAV aerial input images, and the Weighted Myriad Filtering model is utilized for eliminating noise in the first hidden layer. The Hermitian multi-wavelet transform uses the next hidden layer to extract color, texture, and shape features. Later, the Schutz Feature Matching Coefficient classifies UAV images into buildings and non-buildings with higher accuracy and less time. The methodology has demonstrated promising results on a demanding benchmark dataset, significantly reducing time.