Fig. 7

Block diagram to illustrate the proposed circuit component recognition technique designed to facilitate benchmark results on the present datasets. The feature map from DenseNet-121 is passed on to the CBAM attention mechanism. The feature map received from CBAM is flattened using global average pooling (GAP), which serves as an input for the classification layer. The model is trained using the snapshot ensemble, where five snapshots of the model have been saved. The confidence scores of the top-3 snapshots (say, \(CF_3\), \(CF_4\), and \(CF_5\)) undergo a weighted average ensemble technique, where \(w_{i}: i={1, 2, 3}\) is the weight assigned to the confidence score of snapshots to predict the final class label.