A method for identifying fragmentation of open-pit mining blasting based on a new hybrid convolutional neural network

The distribution of rock fragmentation after blasting is an important indicator for assessing the effectiveness of mine blasting. The quantitative characterization of blasting fragmentation is a challenging problem for the evaluation of blasting effects. The use of U-Net network technology to segment blasting images provides a new means for obtaining quantitative statistics from blasting fragmentation. Although the U-Net network is generally capable of segmenting images, there are issues in improving the accuracy and efficiency for ores. To solve these problems, this paper proposes a new network structure - ResOreNet. ResOreNet first integrates the Feature-Fusion module into the U-Net network to become a FU-Net network that enhances the model’s identification of target locations and morphological details, thereby improving the accuracy of ore image segmentation. More specifically, it incorporates the residual network into the FU-Net network, which effectively solves the phenomenon of blurring the boundary of the mineral rock image segmentation caused by the overfitting of the model, and the introduction of the residual network effectively mitigates the problem of the gradient vanishing of the loss function during the backpropagation, and also further improves the computational efficiency of the model, and provide a new technical means for obtaining evaluation indicators of blasting effects in mines.