A Priority-Based Adaptive Firefly Optimized Conv-BiLSTM Algorithm for Electrical Resistance Image Reconstruction

As a high-speed, noninvasive process measurement technology, electrical resistance tomography (ERT) is well suited for visualization of media distribution in industrial and biomedical fields. An innovative optimal hybrid deep-learning strategy utilizing a 1-D convolution neural network (CNN) and recurrent neural network (RNN) is proposed to solve the ERT inverse problem. In the proposed hybrid deep-learning model, the priority-based adaptive firefly algorithm (PAFA) optimizes the neuron structure by feature engineering and adaptively estimates the local optimum to accelerate convergence. The feature selection technique controls randomness through efficient local search and finding the optimal value fit. The input and output relation is enforced by local recurrent cells in the hybrid model feedback with a bidirectional long-short-term memory (BiLSTM) network. The proposed model extracts the latent features from the prior cells during the training period of the network architecture. Simulation and experimental tests are carried out, and the quantitative analysis shows that the proposed hybrid deep-learning model has better imaging accuracy than the traditional image reconstruction methods.