Parallel Fused Dense CNN for Identification of Production from Salt Informatics

Main Article Content

N. Bala Vignesh, Dr. V. Joseph Peter


In this research, a novel Parallel Fused Dense Convolutional Neural Network (PFDCNN) is proposed to extract and identify production features automatically from input salt informatics.  All input salt information is processed through small kernel based densely connected CNN path or phase I and large kernel based densely connected CNN path or phase II. The fused outcomes of these two phases are processed in a Fully Connected (FC) layer to perform one to one connections between input feature map and output class labels. Here, the softmax activation interprets a single vector input features into a number of salt class probabilities. The proposed output class labels are further compared with original class labels from salt dataset for performing an evaluation. Thus, this PFDCNN algorithm achieves 92% of Accuracy, 95% of Recall, 94% of F1-Score and 94% of Precision values, which is 7% of higher accuracy than the existing deep neural network and machine learning methods.

Article Details