计算机技术与发展2026，Vol.36Issue(1)：162-169,8.DOI:10.20165/j.cnki.ISSN1673-629X.2025.0216
基于改进的MobileNetV3的家庭用户用水行为辨识
Household Water Usage Behavior Recognition Based on Improved MobileNetV3
摘要
Abstract
With the deepening development of smart water services in applications such as elderly home care,identifying users'water-usage behaviors has become increasingly important.Ultrasonic water meters,characterized by high measurement accuracy,elevated sampling rates,and digital metering,provide a solid data foundation for behavior recognition.Because inference models deployed on water-meter devices must prioritize power efficiency,the existing MobileNetV3 model is too large and achieves only modest accuracy when recognizing prolonged water-usage activities.We propose an improved lightweight MobileNetV3 model.By streamlining the network architecture and reducing channel counts,the model's size is significantly decreased.Furthermore,we introduce DDSIRB,SK-DDSIRB,and MGGC modules to enhance feature extraction and global information awareness,thereby improving recognition accuracy.Experimental results demonstrate that the enhanced model can accurately identify five basic household water-usage behaviors—dishwasher operation,washing machine use,showering,toilet flushing,and handwashing—from water-meter data.Compared to the original model,it reduces parameter count by 88.65%,compresses model size to 18.87%of the original,and the average F1 score improved by1.45 percentage points,making it better suited for deployment on existing smart water-meter devices.关键词
用水行为辨识/超声波水表/MobileNetV3/模型轻量化/智慧水务Key words
water-usage behavior recognition/ultrasonic water meter/MobileNetV3/model lightweighting/smart water services分类
信息技术与安全科学引用本文复制引用
王晓辉,刘为群,宋可欣,吕方哲,郭丰娟..基于改进的MobileNetV3的家庭用户用水行为辨识[J].计算机技术与发展,2026,36(1):162-169,8.基金项目
中央引导地方科技发展资金(236Z1707G) （236Z1707G）
河北省自然科学基金资助项目(F2022502002) （F2022502002）
