[3B3-E-2] Machine learning: image recognition and generation

[3B3-E-2-05] Sparse Damage Per-pixel Prognosis Indices via Semantic Segmentation

〇Takato Yasuno¹ (1. Research Institute for Infrastructure Paradigm Shift (RIIPS))

キーワード：Deep Learning, Semantic Segmentation, Transfer Learning (FCN, SegNet), Prognosis Indexes, Morphological Image Processing

Efficient inspection and accurate prognosis are required for civil infrastructures with more than 30 years since completion. If we can detect damaged photos automatically per-pixels from the record of the inspection record and countermeasure classification of drone inspection vision, then it is possible that countermeasure information can be provided more flexibly, whether we need to repair and how large the expose of damage interest. A piece of damage photo is often sparse as long as it is not zoomed around damage, exactly the range where the detection target is photographed, is at most only one percent. In this paper, we propose three damage detection methods of transfer learning which enables semantic segmentation in an image with low pixels using damaged photos of drone inspection. Furthermore, we propose prognosis indices to make a decision repair-priority such as the counts index of pop-outs region and the per-pixel area counts index of each pop-out based on morphology image processing. In fact, we show the results applied this method using the 40 drone inspection images whose size is 6,000 x 4,000 on an infrastructure, where each image is partitioned into 400 crops, so the total number of input images is 16,000 for training deep neural network. Finally, future tasks of damage detection modeling are mentioned (211words).