AI-beasd
Analysis Platform
InnoPlatform implemented automatic detection software for abnormal points such as cracks and leaks in concrete structures or facilities using machine learning-based algorithms, and developed a road surface inspection system to secure driving safety of public transport vehicles. In addition, it is focusing on the development of data mining-based decision making technology for time series analysis and risk prediction using the collected measurement data.
Road Condition Detection and Classification using Deep Learning
SSD-MobileNet is a variation of MobileNet, and SSD stands for Single Shot MultiBox Detector (Liu et al., 2015). That is, object detection is performed by SSD and classification is performed by a lightweight CNN, MobileNet. MobileNet is a matrix decomposition model of the convolutional layer for use in mobile devices with limited deep learning performance.
In addition, SSD-Inception-v2 is an object recognition model that combines SSD and Inception-v2 (Loffe and Szegedy, 2015). Inception-v2 is a model in which Batch Normalization (BN) is applied to GoogLeNet (Szegedy et al., 2014) called Inception-v1.
Using the deep learning models, SSD-Inception-v2 and SSD-MobileNet, which learned structural characteristics of each road surface damage type, road damage was detected from ima ge data and the road damage type was classified.
