Infrared image small target detection and identification is a massive challenge with vast applications in essential places such as military industries and airports. Homoplastically, the unclear outline of the target in infrared images, and the complex image background, and the tiny percentage of targets in infrared images make it challenging to accurately acquire the targets’ features and other characteristics, all of which make the research of target detection and identification in infrared images significantly challenging. In this paper, we first contribute an open dataset of Single-frame Infrared Small Target dataset dubbed CAUC-SIRST. After that, the similarity object enhancement module is obtained based on the input feature maps by calculating the Wasserstein distance between the local target and the local background. Then a heterogeneous parallel backbone network structure is constructed, and the feature maps obtained from three different backbone network channels are fused and stitched together. As above, the general convolution channel adaptively extracts feature information; the SimAM channel increases the attention of potential targets, and the similarity object enhancement channel increases the weight of targets. Finally, fusing the feature maps of the three different channels makes it possible to increase the weight of potential targets while retaining the original basic information, this method is named HPN-SOE. Embedding the algorithm into the camera can form a sensor for infrared target detection and recognition. With extensive experimental results demonstrate our outstanding performance, which outperforms other existing methods by achieving a detection accuracy of 85.7% and a detection speed of 31.2 fps.

Solid and accurate object detection in optical remote sensing images still remains significant challenges such as complex background and weak object information. To alleviate above problems, we propose a revolutionary one-stage object detection network. Specifically, the proposed effective localization attention is embedded in deep feature maps with more channels, and is used to locate channels that are effective for detection tasks through one dimensional convolution operations. Following that, a small object compensation strategy is proposed to use compensation fusion operation to enable the reuse and compensation of weak object information. Additionally, a background separation strategy is designed to separate foreground and background, highlighting the features of interest, suppressing background noise. To ascertain our model, extensive experiments are conducted in three public datasets, it can simply achieve mAP of 94.2%, 70.7% and 80.5% in NWPU VHR-10, DIOR and DOTA datasets.

Identifying small objects is essential in airport clearances, such as UAVs and bird flocks affecting the safety of the entire airport. Due to small objects and complex backgrounds, it is a formidable task to learn iconic representations from images. Convolution neural networks (CNNs) is the dominant framework in the past few years. Although CNNs can effectively mitigate local redundancy with convolutional kernels, it is also extremely challenging for convolutional kernels to obtain more feature information about the small object. Alternatively, the prediction box loss function has a critical influence on the eventual model performance. To resolve these problems, a revolutionary small object recognition network is proposed in this paper. In the structure, we propose a hybrid convolutional structure, which is able to obtain object features by different convolutional approaches to achieve more information about the small objects. Following that, we use the Double-One Path Aggregation Network (DO-PAN) feature fusion structure to further integrate the small object information. Finally, we propose all sides consistency intersection over union (AIoU) loss, which enables the model to predict the object location more exactly. Combining the three contributions, we designate this structure as HDA-Net. It can simply achieve 96.1% box mAP50 on the UAV-VIT dataset object recognition task, the inference speed on Tesla T4 is 28.46 fps.

Purpose: This paper aims to propose a new method to solve the two problems in fabric defect detection. Current state-of-the-art industrial products defect detectors are deep learning-based, which incurs some additional problems: (1) The model is difficult to train due to too few fabric datasets for the difficulty of collecting pictures; (2) The detection accuracy of existing methods is insufficient to implement in the industrial field. This study intends to propose a new method which can be applied to fabric defect detection in the industrial field.Design/methodology/approach: To cope with exist fabric defect detection problems, the article proposes a novel fabric defect detection method based on multi-source feature fusion. In the training process, both layer features and source model information are fused to enhance robustness and accuracy. Additionally, a novel training model called multi-source feature fusion (MSFF) is proposed to tackle the limited samples and demand to obtain fleet and precise quantification automatically.Findings: The paper provides a novel fabric defect detection method, experimental results demonstrate that the proposed method achieves an AP of 93.9 and 98.8% when applied to the TILDA(a public dataset) and ZYFD datasets (a real-shot dataset), respectively, and outperforms 5.9% than fine-tuned SSD (single shot multi-box detector).Research limitations/implications: Our proposed algorithm can provide a promising tool for fabric defect detection.Practical implications: The paper includes implications for the development of a powerful brand image, the development of “brand ambassadors” and for managing the balance between stability and change.Social implications: This work provides technical support for real-time detection on industrial sites, advances the process of intelligent manual detection of fabric defects and provides a technical reference for object detection on other industrialOriginality/value: Therefore, our proposed algorithm can provide a promising tool for fabric defect detection.

Unmanned aerial vehicles (UAV) “black flight” incidents cause serious security risks and economic losses to airports. Additionally, existing UAV detection methods are mainly radar technology at airports. Whereas, it is unable to correctly identify the number of UAVs and visualise their size, which undoubtedly poses a serious security risk. Accordingly, an anchor‐free UAV detection method HollowBox is proposed to supplement radar detection equipment for the airport “black flight” problem. It is inspired by the FoveaBox object detection method, the object detection feature layers are reset and the allocation ratio of positive and negative samples in the training phase are redefined, the HollowBox UAV detection is proposed according to the multi‐size characteristics. Extensive experiments show that, the approach achieves 90.1% AP, 6% false detection rate and 17.2 FPS inference speed, which accomplished a satisfactory performance, as verified by the real‐shot data collected of an airport in Tianjin. This work is of great significance for the application in airport UAV detection.

Existing research has demonstrated that meta-learning methods hold considerable promise in addressing the challenges posed by few-shot object detection. However, remote sensing scenarios present two major challenges. The sparse features of small objects provide insufficient support information for query enhancement, and significant morphological variations caused by lighting and viewpoint differences hinder intra-class consistency capture via direct alignment in few-shot learning. To address these challenges, we propose a generative meta-learning detection framework. The framework first introduces a Dynamic Relation Dual-Stream Network to achieve dynamic support-query feature alignment through joint modeling of evolutionary and relational features, thereby enhancing representation in few-shot conditions. Second, an Optimal Transport-based Generative Meta-Learner is developed to mitigate feature distribution bias via generative augmentation in latent space. Additionally, an Orthogonal Frequency Decomposition Head is incorporated to adaptively separate query features into low-frequency contour and high-frequency detail components, effectively suppressing background noise interference. Experiments on multiple remote sensing datasets demonstrate that the proposed method achieves consistent performance gains over leading baseline methods in various few-shot settings. Its effectiveness is further validated across different backbone networks, highlighting strong generalization in few-shot remote sensing object detection.