Detecting Building Anomalies by Applying an Innovative CNN-Based Approach in the Casablanca Region Using Aerial Imagery

Abstract

Globally the growing urbanization of Morocco, particularly in Zenata City, is giving rise to environmental problems. For the development of Zenata, as a smart city, and modern urban land management, keeping an eye on the quantity and location of construction is essential. The purpose of this study was to use drone imagery and a yolov8 CNN-based object detection algorithm to detect constructions and irregular constructions. In this study, an orthophoto of Zenata city is divided into 742 georeferenced images and were separated by three distinct datasets, training (63%), testing (22%) and validation (15%). The yolov8 model was trained using the training sets that were manually labeled on construction images in the training and validation sets. The network's performance was assessed using training and validation datasets. According to the performance findings, the system trained for only three 1.600 hours, reaching its maximum accuracy at epoch 300. The model's average precision was 87-98%. Yolov8 can detect construction using images from drones with exceptional accuracy and computing performance.