Millimeter-wave (mmWave) and terahertz (THz) communications are susceptible to frequent link disruptions and severe performance degradation due to high directionality, significant path loss, and sensitivity to blockages. These challenges are particularly acute in highly dynamic and densely populated user environments. The issues present significant obstacles to ensuring reliability and quality of service (QoS) in future space–air–ground integrated networks. To address these challenges, this paper proposes an adaptive transmission control scheme designed for space–air–ground integrated multi-hop networks operating in the mmWave/THz bands. By analyzing the intermittent connectivity inherent in such networks, the proposed scheme incorporates an incremental factor and a backlog indicator into its congestion control mechanism. This allows for the accurate differentiation between packet losses resulting from network congestion and those caused by channel blockages, such as human body occlusion or beam misalignment. Furthermore, the scheme optimizes the initial congestion window during the slow-start phase and dynamically adapts its transmission strategy during the congestion avoidance phase according to the identified cause of packet loss. Simulation results demonstrate that the proposed method effectively mitigates throughput degradation from link blockages, improves data transmission rates in highly dynamic environments, and sustains more stable end-to-end connectivity. Our proposed scheme achieves a 35% higher throughput than TCP Hybla, 40% lower latency than TCP Veno, and maintains 99.2% link utilization under high mobility.

The rapid advancement of digital technology has enhanced medical image transmission and storage; yet network openness and anonymity pose risks to copyright and privacy. Given the limitations of existing zero-watermarking schemes in dealing with complex geometric attacks, we propose a robust zero-watermarking scheme that leverages a residual-based GAN (Res-GAN). Our scheme enhances the security of the watermark image by scrambling it with a logistic chaotic map and extracts robust features from medical images by harnessing a pretrained Res-GAN model. These features are converted into a binary vector via the average hash algorithm, which is then XORed with the scrambled watermark image to generate the zero-watermark. Experimental results highlight that integrating residual modules into GAN boosts the scheme’s robustness against geometric attacks. Under various attacks, our scheme can still maintain the normalized correlation coefficient values above 0.84. Compared with other existing zero-watermarking schemes for medical images, our scheme has stronger robustness. This scheme offers effective technical support and practical guidance for medical image copyright and privacy protection. It enhances the security of medical image transmission and sharing, fostering data circulation and cooperation. As medical informatization advances, this technology will be pivotal in medical image management, telemedicine, and big data analysis.

The license plate angle is unfixed, the vehicle position is ununiform, and the picture is not sufficiently illuminated which leads to the decrease of license plate recognition accuracy. In order to improve the accuracy of license plate recognition, a deep learning-based license plate recognition method is proposed. For license plate location problem, YOLOv3 algorithm is used. The algorithm is more capable of recognizing small targets and is suitable for license plate location recognition that requires precise positioning. For the problem of license plate character recognition, the CNN plus multitask recognition method is proposed for recognition. The experimental results show that the accuracy of the license plate recognition method proposed in this paper reaches 96%, and the intelligent license plate recognition is realized.

This paper mainly explores the method of Chinese cursive character recognition, establishes the Standard Cursive database in the process of research, and puts forward the similarity distance to measure the similarity between the cursive font to be recognized and the character set, and improves the calculation method of the similarity distance. Through experimental comparison, pH algorithm performs best in cursive character recognition.