Enhanced dense connectivity network model applied to ECT image reconstruction

In order In solve tint problem tliat traditional neural networks cannot integrate the bottom position features ant! the top semantic features of the capacitive feature tensor well in ECT image reconstruction, an enhanced it-use connection network model was prop<ned. First, tlu- initial diek-ctric constant distribution was obtained hy training tin; Fully Connected Neural Network, anil tlu- output characteristic map of tlu- fully connected neural network was used as tin- input of tin-compensated L-net network. Sceorully. a compensated L-net network was built, atul a DenseNel-like dense jump connection utcclianism was added Ix-tween I he encoder anil decoder to n*lain a large amount of underlying loealion feature information and reduce the feature loss of multiple output nodes of tlu- nuidel. At tlie same lime, (lie mulli-seale ifc-nse eavity iron vol utional module was used in replace tin- onlinary convolution in tlu- compensated U-net to enlarge llu- receptive field of llie model ami enrich tlu- multi-scale infonnahon. Finally, an efficient channel attention mechanism module was used to realize the cross-channel interaction of tin- output features of sub-decoder nodes, which enhances tin- model s attention lo imjiortant information and improves the nonlinear fitting ability of the model. Tlu- experimental results show lliat tlie reconstructed images based on this algorithm have higher resolution, clearer imaging edges, and more robustness I ban the 1-indwelier iterative algorithm and tlu- L-net algorithm. © 2024 Chinese Vibration Engineering Society. All rights reserved.