This paper addresses two vital challenges in Unsupervised Domain Adaptation (UDA) with a focus on harnessing the power of Vision-Language Pre-training (VLP) models. Firstly, UDA has primarily relied on ImageNet pre-trained models. However, the potential of VLP models in UDA remains largely unexplored. The rich representation of VLP models holds significant promise for enhancing UDA tasks. To address this, we propose a novel method called Cross-Modal Knowledge Distillation (CMKD), leveraging VLP models as teacher models to guide the learning process in the target domain, resulting in state-of-the-art performance. Secondly, current UDA paradigms involve training separate models for each task, leading to significant storage overhead and impractical model deployment as the number of transfer tasks grows. To overcome this challenge, we introduce Residual Sparse Training (RST) exploiting the benefits conferred by VLP’s extensive pre-training, a technique that requires minimal adjustment (approximately 0.1%~0.5%) of VLP model parameters to achieve performance comparable to fine-tuning. Combining CMKD and RST, we present a comprehensive solution that effectively leverages VLP models for UDA tasks while reducing storage overhead for model deployment. Furthermore, CMKD can serve as a baseline in conjunction with other methods like FixMatch, enhancing the performance of UDA. Our proposed method outperforms existing techniques on standard benchmarks. Our code will be available at: https://github.com/Wenlve-Zhou/VLP-UDA.

In recent years, impressive performance of deep learning technology has been recognized in synthetic aperture radar (SAR) automatic target recognition (ATR). Since a large amount of annotated data are required in this technique, it poses a trenchant challenge to the issue of obtaining a high recognition rate through less labeled data. To overcome this problem, inspired by the contrastive learning, we proposed a novel framework named batch instance discrimination and feature clustering (BIDFC). In this framework, different from that of the objective of general contrastive learning methods, embedding distance between samples should be moderate because of the high similarity between samples in the SAR images. Consequently, our flexible framework is equipped with adjustable distance between embedding, which we term as weakly contrastive learning. Technically, instance labels are assigned to the unlabeled data in per batch, and random augmentation and training are performed few times on these augmented data. Meanwhile, a novel dynamic-weighted variance loss (DWV loss) function is also posed to cluster the embedding of enhanced versions for each sample. The experimental results on the moving and stationary target acquisition and recognition (MSTAR) database indicate a 91.25% classification accuracy of our method fine-tuned on only 3.13% training data. Even though a linear evaluation is performed on the same training data, the accuracy can still reach 90.13%. We also verified the effectiveness of BIDFC in OpenSarShip database, indicating that our method can be generalized to other data sets. Our code is available at: https://github.com/Wenlve-Zhou/BIDFC-master.

Hand gestures are a well-known and straightforward method of human-computer interaction. The majority of the study focused on hand gesture recognition. However, little work has been done to develop a complete set of gesture recognition applications. With the improvement of model feature extraction ability and the increase in the number of model parameters, it is becoming more challenging to achieve a small memory footprint on mobile devices based on an ARM architecture or CPU devices based on x86 architecture. However, these existing methods are heavy, requiring more memory and inference time. The execution of memory-efficient CNNs without compromising accuracy has been a challenge, especially when the inference has to be performed on an edge computing device in real time. Therefore, we propose a lightweight network for hand gesture recognition (LHGR-Net) and deploy it on a Raspberry Pi. LHGR-Net consists of three main parts: the base network structure, the multiscale structure (MSS), and the lightweight attention structure (LAS). We present pre-trained weights that are learned from other data to initialize the network structure. In addition, the LHGR-Net model was made to be deployed on a Raspberry Pi, and a deployed model can be used to control home appliances. Extensive experiments show that our method achieves almost as good as state-of-the-art performance in hand gesture recognition and running time.

Zero-shot remote sensing image scene classification (ZS-RSISC) aims to identify remote sensing (RS) image scenes of unseen classes whose samples are unavailable in the training stage. To transfer knowledge from seen RS classes to unseen RS classes, existing methods either rely on laborious manual labeling to learn semantic features or directly use the word embeddings learned based on general corpus and independently of zero-shot models. They ignore the complex inter-class correlation information which plays a vital role in communicating seen with unseen classes. Besides, current studies in ZS-RSISC impose the same penalty to equally constrain each class for the inter-class separation and intra-class compactness, which results in unclear classification boundaries. In this paper, we tackle ZS-RSISC via graph-based semantic embedding refinement (GSER) in an end-to-end manner. We propose semantic graph convolutional networks (S-GCNs) to explore the correlation structure among classes in a unified framework. The semantic graph embeddings are further refined by the learning of the semantic-guided class patterns and component patterns. Specifically, we propose adaptive additive separation (AAS) loss to adaptively adjust appropriate penalty for each class and explicitly promote intra-class compactness and inter-class separation. Further, instance-level alignment and class-level alignment are proposed to enhance the discriminative ability of the semantic-guided class patterns. To alleviate model bias towards seen classes, semantic-guided component patterns shared by seen and unseen classes are exploited via feature reconstruction. Extensive experiments on both the zero-shot and generalized zero-shot settings demonstrate the effectiveness of our proposed GSER.

As samples of steel defects are industrially limited, it is challenging for most deep learning methods that rely on ample labeled data to identify steel surface defects. Recently, contrastive learning has achieved good performance in natural image classification tasks with few labeled samples, yet two obstacles prevent its effective application to steel surface defect images. One is that due to the presence of inter-class and intra-class similar samples in steel surface defect, the fixed contrast strength in contrastive learning will destroy the potential semantic information of defect samples. Another is that contrastive learning requires a large amount of unlabeled data, whereas steel surface defect samples are insufficient. To overcome the above-mentioned problems, a novel framework named flexible and diverse contrastive learning (FDCL) is proposed. This framework consists of two parts, flexible contrast (FiCo) and diverse generative adversarial networks (DGANs). Diverse images generated by the DGAN and real images are fed into FiCo for representation learning. In the FiCo, the contrast strength among samples is flexibly adjusted by the proposed variable temperature discrimination and feature reconstruction (FR). In addition, the output features (OF) of FiCo will be used as input to the DGAN generator to improve image quality, thus further facilitating representation learning. The proposed FDCL is implemented on four standard steel surface defect data sets, and the experimental results demonstrated that it achieves superior performance over state-of-the-art methods. Our code is available at: https://github.com-/jiacongc/FDCL .

In the field of measuring parameters of mobile communication base station antenna, most of its methods share some deficiencies to a different extent. The traditional ones suffer from inferior efficiency and low safety coefficient while the state-of-the-art ones reveal problems of unsatisfying operational speed and accuracy. To address these problems, a high-speed and high-precision measuring system has been proposed with AntennaNet to speed up instance segmentation. First, the dual attention mechanism module is designed and adopted in AntennaNet, which can promote the accuracy and efficiency of the detection and segmentation of mobile base station antenna, while the PointRend module with down-tilt fitting is also employed to smooth the edges of masks segmented from antenna images and measure the parameters data. Second, transfer learning is introduced to ameliorate overfitting and training difficulty caused by small samples. Third, threshold, pixel coordinates, and least square are applied to ascertain the number of antennas and measure the relevant parameters. Finally, this high speed and high precision measuring system is complete with the combination of android App and UAV. Experimental results fully attest to the preponderance of the proposed method in its significant improvement of segmentation speed, fitting accuracy speed, and fitting accuracy of the antenna, whose relevant parameters of the fitting also conform to industry standards. Our method not only ensures staff safety but also improves social efficiency, which revolutionized the way of measuring antenna parameters of mobile communication base stations.