Spatial Distribution of Blast Reactor Block Based on TAUNet Segmentation Model
In order to better meet the need for real-time and high-precision detection of blast reactor block in mining sites,a blast reactor block segmentation model TAUNet(Transformer Aspp UNet)based on deep learning was proposed.The model integrated Transformer's self-attentive mechanism in the encoder and decoder of UNet,used it to handle large feature mappings,improved the extraction of global information and restored the granularity details skipped in the encoder.In the backbone network feature extraction stage,the ASPP null convolution module was incorporated to enhance the model for block local feature fusion.On the basis of the blast reactor image segmentation,the spatial distribution information of the blast reactor block was obtained by the method of layering blast reactor.The results show that the TAUNet segmentation model has a good segmentation performance,and the model training evaluation indexes,including dice coefficient,intersection over union,and recall rate reach 97.12％、94.61％and 96.2％respectively,which are all better than the mainstream semantic segmentation model.And the TAUNet model have good segmentation effects on the on-site blast reactor blocks.Through the method of layering blast reactor,the spatial distribution of blast reactor block of 315-300 m platform in the west mining area of a mine in Zhaoqing City is determined as 87.15％of the block size distribution in the 0-0.6 m,9.9％of the block size distribution in the 0.6-1.0 m,and large blocks with size greater than 1.0 m accounted for 2.95％.The research results can provide a reference for the refined and intelligent development of blasting effect evaluation.
万方数据
维普
