Labelme的問題，透過圖書和論文來找解法和答案更準確安心。 我們找到下列免費下載的地點或者是各式教學

Labelme進入發燒排行的影片

Automated Vision-based Post-Earthquake Safety Assessment for Bridge Using STF-PointRend and EfficientNetB0

為了解決Labelme的問題，作者Kenneth Harsono 這樣論述：

Structural health monitoring (SHM) on the bridge is important to know the usability of the bridges. However, conventional inspection is labor-intensive and expensive. This method is not suitable for post-earthquake inspections that require speed and consistency. Therefore, this research aims to dev

elop an automated bridge inspection using STF-PointRend and EfficientNetB0. The STF-PointRend consists of two-part, namely symbiotic organism search as a hyper-parameter optimizer and PointRend as semantic segmentation. This model is used to recognize the component and the damage type which will be

used to get the percentage of the damaged component. On the other hand, the EfficientNetB0 uses as the image classifier. The output of this model is used to get the damage level from each component. As a base to determine the safety of the bridge, this study uses the degree of earthquake resistance.

This rating system is based on the DERU method but only considers the structural component. The result shows that STF-PointRend gets a good testing result with the mIoU of 82.67% and 71.42% for component and damage detection. Meanwhile, the EfficientNet got an average F1score of 0.85912 for the tes

ting dataset. For further evaluation, this research uses two minor bridges that suffered catastrophic earthquakes from Palu Earthquake in 2018. The evaluation shows that both bridges need maintenance as soon as possible.

Computer Vision-Based Post-Earthquake Inspections on Building Safety Assessment

為了解決Labelme的問題，作者Alvin Kwek 這樣論述：

Safety assessment in structural health monitoring is one of the important factors for buildings. Structural health monitoring for building inspection is a necessity to provide faster danger responses and reduce threats. However, building infrastructures were exposed to natural disaster and lifetime

. For the safety assessment during disaster, the image that were taken by the expert urgently from the site does not always produce a good quality image. Therefore, the Hybrid-GAN which comprise of ESRGAN and DeblurGANv2 were used as the image repairing method that generating better images. Meanwhil

e, Deep Learning techniques of Convolutional Neural Network (CNN) usage has significantly increase in the modern days especially computer vision-based approach for safety assessment in semantic segmentation task. Moreover, this research proposed Transfer Learning U-Net (TF-Unet) Algorithm to detect

and classify building components which are column and structural wall alongside with the damage level according to the Taiwan codes for building evaluation assessment. Furthermore, the pre-train model were used to predict three case studies to test the capability of TF-Unet in handling real word dat

aset.