| Abstrakt: |
Localization Wireless Capsule Endoscopy (WCE) inside the human small intestine is a hard issue for a decade. This is due to long, curly, and compact structure small intestine. Some of the techniques as Radio Frequency (RF), Vision based and Magnetic type have been proposed. To be more important, any one of the techniques as RF, Vision or Magnetic shows that the poor performance in terms of localization error and accuracy. To address these issues, in this paper a hybrid RF with Vision aware Fusion scheme (RF-VaF) is proposed under multisensor. In RF based approach, Time of Flight and Received Signal Strength Indicator are presented. In vision based approach, Siamese CapsNet is proposed for frames registration, correlation maps generation, and pixel based matching point's prediction. A multi-feature extraction (color, edge, intensity and texture) is executed by Spatial Transformer Network for consecutive frames. In particular, this will be fed into the Siamese CapsNet. Similarly, Canberra distance is computed in the softmax layer for localization. The results from RF and Vision are fused into find the accurate position. In this step, hydrological cycle optimization algorithm is proposed. With this step, WCE can be accurately predicted at the end. One of the novel steps here is adjusting the Receiver's Position by Positioning Metric. Finally, the performance is computed by using Matlab R2019b. From the results, it is proved that the RF-VaF is outperforms than the previous works by following metrics as Average Localization Error [5.41], Root Mean Square Error [6.76], Normalized Error [6.775], Localization Accuracy [96.43%], Localization Error [5.14%], Sensitivity [96.6%] and also Specificity [96.5%]. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink