WLD-InceptionV3: AI and Computer Vision-Based Urban Waterlogging Detection Using a Modified InceptionV3 Model

Urban waterlogging monitoring is now crucial for urban management due to its frequent occurrence.Early and accurate detection of floods has now become crucial to reduce waterlogging impacts in urban areas.This research explores the adaptability of deep learning techniques, specifically the InceptionV3 model, for urban flood detection.This paper helps to differentiate between the deep learning models VGG16, InceptionV3, Resnet50, and MobileNet and their performance for flood detection tasks.InceptionV3 is chosen for its effectiveness in transfer learning with extracted features from pre-trained ImageNet layers.The model architecture was modified by including global average pooling and a dropout layer with a dense output layer.Fine-tuning is performed in its last few layers to specifically train the urban flooding images.This research demonstrates the potential use of deep learning models in environmental disaster detection tasks like flood detection.It also demonstrates the power of a deep learning model like InceptioV3 in image recognition and classification tasks with transfer learning techniques with exceptional performance with 98.50% accuracy, an F1 score of 0.98, and precision-recall scores of 0.96 and 1.0, respectively.This flood detection system has the potential for real-time deployment, analyzing images from CCTV cameras, mobile phones, or publicly available sources.Real-time flood detection can trigger automatic alerts, notifying residents and authorities before disasters strike.This is particularly relevant considering the frequent post-rain flooding events in many urban areas.A practical application could involve integrating the model with traffic-monitoring CCTV cameras in busy areas.Upon detecting floods, the system