In recent years a vast amount of visual content has been generated and shared from many fields, such as social media platforms, medical imaging, and robotics. This abundance of content creation and sharing has introduced new challenges, particularly that of searching databases for similar content — Content Based Image Retrieval (CBIR) — a long-established research area in which improved efficiency and accuracy are needed for real-time retrieval. Artificial intelligence has made progress in CBIR and has significantly facilitated the process of instance search. In this survey we review recent instance retrieval works that are developed based on deep learning algorithms and techniques, with the survey organized by deep feature extraction, feature embedding and aggregation methods, and network fine-tuning strategies. Our survey considers a wide variety of recent methods, whereby we identify milestone work, reveal connections among various methods and present the commonly used benchmarks, evaluation results, common challenges, and propose promising future directions.

In recent years, analysis and optimization algorithm based on image data is a research hotspot. Aircraft detection based on aerial images can provide data support for accurately attacking military targets. Although many efforts have been devoted, it is still challenging due to the poor environment, the vastness of the sky background, and so on. This paper proposes an aircraft detection method named TransEffiDet in aerial images based on the EfficientDet method and Transformer module. We improved the EfficientDet algorithm by combining it with the Transformer which models the long-range dependency for the feature maps. Specifically, we first employ EfficientDet as the backbone network, which can efficiently fuse the different scale feature maps. Then, deformable Transformer is used to analyze the long-range correlation for global feature extraction. Furthermore, we designed a fusion module to fuse the long-range and short-range features extracted by EfficientDet and deformable Transformer, respectively. Finally, object class is produced by feeding the feature map to the class prediction net and the bounding box predictions are generated by feeding these fused features to the box prediction net. The mean Average Precision (mAP) is 86.6%, which outperforms the EfficientDet by 5.8%. The experiment shows that TransEffiDet is more robust than other methods. Additionally, we have established a public aerial dataset for aircraft detection, which will be released along with this paper.

Deep learning for fine-grained image retrieval in an incremental context is less investigated. In this paper, we explore this task to realize the model’s continuous retrieval ability. That means, the model enables to perform well on new incoming data and reduce forgetting of the knowledge learned on preceding old tasks. For this purpose, we distill semantic correlations knowledge among the representations extracted from the new data only so as to regularize the parameters updates using the teacher-student framework. In particular, for the case of learning multiple tasks sequentially, aside from the correlations distilled from the penultimate model, we estimate the representations for all prior models and further their semantic correlations by using the representations extracted from the new data. To this end, the estimated correlations are used as an additional regularization and further prevent catastrophic forgetting over all previous tasks, and it is unnecessary to save the stream of models trained on these tasks. Extensive experiments demonstrate that the proposed method performs favorably for retaining performance on the already-trained old tasks and achieving good accuracy on the current task when new data are added at once or sequentially.

With the rapid growth of application demands and the real-time change of environmental situations, the defects of the UAV task network in adaptability, flexibility, and resilience are becoming more and more prominent. The current network architecture that the junction of points and lines is fixed cannot dynamically provide capacity requirements in real-time due to the failure nodes encountered in the Unmanned Aerial Vehicle (UAV) task scheduling process. To address this challenging issue, this paper proposes a flexible network architecture supporting dynamic fault-tolerant task scheduling model (DSM-FNA) for the UAV cluster. To be specific this paper resorts to super network theory, combining the management theory of flexible network and resilience network to carry out the organizational calculation on the model, and also draw upon linear transformation function to weight and stratify the capability value according to the ability requirement required by the task. Then, a flexible network architecture dynamic scheduling algorithm (FDSA) is proposed, and the substitution strategy is designed for the failure point, to realize the capability and dynamically adapt to the task. Finally, compared with the classical Max-Min algorithm and other algorithms, it is verified that the FDSA algorithm performs better dynamic adjustment for quick response in case of UAV cluster emergencies.

Indoor positioning based on existing Wi-Fi fingerprints is becoming more and more common. Unfortunately, the Wi-Fi fingerprint is susceptible to multiple path interferences, signal attenuation, and environmental changes, which leads to low accuracy. Meanwhile, with the recent advances in charge-coupled device (CCD) technologies and the processing speed of smartphones, indoor positioning using the optical camera on a smartphone has become an attractive research topic; however, the major challenge is its high computational complexity; as a result, real-time positioning cannot be achieved. In this paper we introduce a crowd-sourcing indoor localization algorithm via an optical camera and orientation sensor on a smartphone to address these issues. First, we use Wi-Fi fingerprint based on the K Weighted Nearest Neighbor (KWNN) algorithm to make a coarse estimation. Second, we adopt a mean-weighted exponent algorithm to fuse optical image features and orientation sensor data as well as KWNN in the smartphone to refine the result. Furthermore, a crowd-sourcing approach is utilized to update and supplement the positioning database. We perform several experiments comparing our approach with other positioning algorithms on a common smartphone to evaluate the performance of the proposed sensor-calibrated algorithm, and the results demonstrate that the proposed algorithm could significantly improve accuracy, stability, and applicability of positioning.

Knowledge graph (KG) embedding for predicting missing relation facts in incomplete knowledge graphs (KGs) has been widely explored. In addition to the benchmark triple structural information such as head entities, tail entities, and the relations between them, there is a large amount of uncertain and temporal information, which is difficult to be exploited in KG embeddings, and there are some embedding models specifically for uncertain KGs and temporal KGs. However, these models either only utilize uncertain information or only temporal information, without integrating both kinds of information into the underlying model that utilizes triple structural information. In this paper, we propose an embedding model for uncertain temporal KGs called the confidence score, time, and ranking information embedded jointly model (CTRIEJ), which aims to preserve the uncertainty, temporal and structural information of relation facts in the embedding space. To further enhance the precision of the CTRIEJ model, we also introduce a self-adversarial negative sampling technique to generate negative samples. We use the embedding vectors obtained from our model to complete the missing relation facts and predict their corresponding confidence scores. Experiments are conducted on an uncertain temporal KG extracted from Wikidata via three tasks, i.e., confidence prediction, link prediction, and relation fact classification. The CTRIEJ model shows effectiveness in capturing uncertain and temporal knowledge by achieving promising results, and it consistently outperforms baselines on the three downstream experimental tasks.