By considering the different cumulant combinations of the 2FSK, 4FSK, 2PSK, 4PSK, 2ASK, and 4ASK, this paper established new identification parameters to achieve the recognition of those digital modulations. The deep neural network (DNN) was also employed to improve the recognition rate, which was designed to classify the signal based on the distinct feature of each signal type that was extracted with high order cumulants. The extensive simulations demonstrated the exceptional classification performance for new key features based on high order cumulants. The overall success rate of the proposed algorithm was over 99% at the signal to noise ratio (SNR) of −5 dB and 100% at the SNR of −2 dB. The results of the experiments also showed the robustness of the proposed method for a variety of conditions, such as frequency offset, multi-path, and so on. INDEX TERMS Modulation recognition, high order cumulants, deep learning, wireless communications.

No abstract

No abstract

Semantic segmentation has been continuously investigated in the last ten years, and majority of the established technologies are based on supervised models. In recent years, image-level weakly supervised semantic segmentation (WSSS), including single-and multi-stage process, has attracted large attention due to data labeling efficiency. In this paper, we propose to embed affinity learning of multi-stage approaches in a single-stage model. To be specific, we introduce an adaptive affinity loss to thoroughly learn the local pairwise affinity. As such, a deep neural network is used to deliver comprehensive semantic information in the training phase, whilst improving the performance of the final prediction module. On the other hand, considering the existence of errors in the pseudo labels, we propose a novel label reassign loss to mitigate over-fitting. Extensive experiments are conducted on the PASCAL VOC 2012 dataset to evaluate the effectiveness of our proposed approach that outperforms other standard single-stage methods and achieves comparable performance against several multi-stage methods. CCS CONCEPTS• Computing methodologies → Scene understanding.

In recent years, more and more deep learning methods for fault diagnosis of rolling element bearings (REBS) have been proposed. However, in industry, the scarcity of available data to monitor the health condition of REBS leads to a low recognition accuracy of the trained intelligent diagnostic models. To solve this problem, we propose a simulation data driven subdomain adaption adversarial transfer learning network (SAATLN). Firstly, a defect vibration model is introduced to simulate vibration signals of different types of REBS faults. And the real signal and simulated signal are used as the target domain and source domain of the transfer learning fault diagnosis methods, respectively. Secondly, SAATLN uses the designed residual Squeeze-and-Excitation (Re-SE) blocks to extract transfer features between different domains. Meanwhile, it combines adversarial learning and subdomain adaptation to adapt the marginal distribution and conditional distribution discrepancy of high-level features. And the local maximum mean discrepancy (LMMD) is introduced as the subdomain adaptation metric criterion. Finally, different transfer tasks are performed on the artificially damaged and run-to-failure REBS data sets. The results demonstrate the effectiveness and superiority of the SAATLN in the simulation data driven REBS fault diagnosis.

No abstract

No abstract

No abstract