Deep learning is currently the mainstream method of object detection. Faster region-based convolutional neural network (Faster R-CNN) has a pivotal position in deep learning. It impressive detection effects ordinary scenes. However, under special conditions, there can still be unsatisfactory performance, such as having problems like occlusion, deformation, or small size. This paper proposes novel and improved algorithm based on R-CNN framework combined with skip pooling fusion contextual...

Fine-grained ship detection is an important task in high-resolution satellite remote sensing applications. However, large aspect ratios and severe category imbalance make fine-grained a challenging problem. Current methods usually extract square-like features that do not work well to detect ships with ratios, the misalignments feature representation will severely degrade performance of localization classification. To tackle this, we propose shape-aware learning method mitigate during...

Few-shot object detection (FSOD) eliminates the dependence on tremendous instances with manual annotations in conventional detection. We deem that scarcity of positive samples is main reason restricts performance FSOD detectors. In this paper, a novel model via sample processing, namely, FSSP, proposed to detect objects accurately only few annotated samples, which based structural design Siamese network and uses YOLOv3-SPP as baseline. Central FSSP are our designed self-attention (SAM)...

Fast and accurate ship target detection technology plays an important role in improving driving safety, rescue at sea, marine environmental protection, sea traffic control. It is also one of the key technologies for development informatization intelligence. However, current models used different scales multiple scenarios exhibit high complexity slow inference speed. The trade-off between model speed accuracy limits deployment on edge devices. This study proposes a lightweight multi-scale...

Ship detection in satellite images is a challenging task. In this paper, we introduce transfer learned Single Shot MultiBox Detector (SSD) for ship detection. To end, state-of-the-art object model pre-trained from large number of natural was with limited labeled images. the best our knowledge, could be one first studies which SSD into on Experiments demonstrated that method achieve 87.9% AP at 47 FPS using NVIDIA TITAN X. comparison Faster R-CNN, 6.7% improvement achieved. Effects...

Detecting airplanes from high-resolution remote sensing images has a variety of applications. The characteristics clear details, rich spatial and texture information objects in make it possible to identify different types backgrounds. However, usually exhibit slight inter-class discrepancy unbalanced class distribution, which pose significant challenges fine-grained detection airplanes. In this paper, we propose the ISCL, an Instance Switching-based Contrastive Learning method for airplane...

Weakly supervised object detection aims at learning detectors with only image-level category labels. Most existing methods tend to solve this problem by using a multiple instance detector which is usually trapped discriminate parts, rather than the entire object. In order select high-quality proposals, recent works lever-age objectness scores derived from weakly-supervised segmentation maps rank proposals. Base our observation, kind of guided method always fails due neglect fact that all...

Military object detection technology plays an important role in realizing the informatization and intelligence of military equipment, but complex environment scarce sample data objects also become difficulties detection. This paper proposes a framework based on optimal Gabor filtering deep feature pyramid networks. At first, we combine texture characteristics with requirements tasks proposed Fine Region Proposal Network (FRPN). A set filter is designed screened this scheme, construct Banks...

Due to the development of deep learning, in recent years, field computer vision grows rapidly. A large amount technologies have been applied actual problems. At present, industry vehicle damage assessment requires a lot manpower, and new automatic intelligent technology can greatly reduce industrial costs. In this paper, framework algorithm based on object detection image classification is proposed. This automatically identify position, type degree according photos provided by users, so as...

Aiming at the problems of low extraction accuracy, high missed detection rate and learning efficiency existing behaviour tracking analysis models, a model MOOC online education platform based on XAPI Bayesian fuzzy rough set is established. Firstly, stratified, then its correlation with effect are analysed, indicators determined. Finally, The experimental results show that accuracy always above 93%, high; missing between 1% 4%, low; maximum improvement 14.8%, students' high, which verifies...

In this paper, a novel despeckling method based on Gaussian scale mixtures (GSM) model in the directionlet domain is proposed. Before despeckling, we define measurement of directivity texture to calculate according edge map. After transform, neighborhoods coefficients at adjacent scales are modeled as GSM model. Under model, Bayes Least Squares (BLS) estimator adopted reduce speckle noise. Quantitative and qualitative experimental results show that proposed an effective tool for SAR images....

Abstract The popular convolutional neural networks (CNN) require data augmentation to achieve rotation invariance. We propose an alternative mechanism, Pre-Rotation Only at Inference stage (PROAI), make CNN invariant. overall idea is learn how the human brain observe images. At training stage, PROAI trains a with small number using images only one orientation. inference introduces pre-rotation operation rotate each test image into its all-possible orientations and calculate classification...

Abstract It is crucial for vehicular communications to optimize the field strength coverage on roads, which can be illustrated by radio map (RM). In this paper, a deep convolutional neural network‐long short‐term memory (DCNN‐LSTM) model construction of road RM proposed. First, multi‐modal dataset built, including measured strength, longitude, latitude and elevation data obtained at various measurement points, as well outline maps buildings that are close points. Second, DCNN‐LSTM designed...

An improved wavelet thresholding method is applied to denoise weather radar raw data. It used an adjustable other than soft or hard methods. Comparing with median filtering method, the result of this shows that it can remove noise from data and keep useful infomation well at same time. improve detection rate lower false alarm for small scale sever storm. The echo maps are also made clearer.

In order to solve the problem of model accuracy reduction caused by difficulty obtaining specific training samples or insufficient number in application existing object detection and recognition based on deep learning, this article proposes a conditional generative adversarial network (VSA-CGAN), which integrates self-attention mechanism visual perception optimize inference attention feature maps, so as learn global information image detailed features object. It is designed add generator...

Detection of targets on sea surfaces is an important area application that can bring great benefits to the management and control systems in marine environments. However, there are few open-source datasets accessible for purpose object detection seas rivers. In this paper, a study conducted improved algorithms based YOLOv5 model. The dataset tests contains ten categories typical objects commonly seen contexts seas, including ships, devices, structures. Multiple augmentation methods employed...

A fast variational fusion model based on partial differential equation (PDE) is presented for pan-sharpening. The functional constructed with several energy terms. gradient term created by calculating the vector field of panchromatic image. visualization used improving perceptual effect and spectral preserving designed enforcing coherence. channel correlation radiometric reduction are defined to preserve multi-spectral channels decrease distortion. Inspired shock-filtering model, an inverse...

Multi-legged rescue equipment plays an important role in emergency rescue, military reconnaissance and due to its flexibility adaptability. The ability of terrain recognition scene segmentation is guarantee for the robot surmount obstacles automatically, as part it, point cloud semantic has also been greatly developed recent years. However, existing methods are all urbanization scenes or indoor objects, field relatively vacant. paper aims achieve real-time equipments. First, rule-based...

With the rapid progress of computational science and computer simulation ability, a lot properties can be predicted by powerful ability parallel computation before actual research development.With development high performance architecture, GPU is more widely used in field as an emerging growing number computations use heterogeneous cluster architecture.However, how to partition workload map computing resource has always been focus difficult point.In current study GPU, according problems...

Aiming at the problems of low extraction accuracy, high missed detection rate and learning efficiency existing behaviour tracking analysis models, a model MOOC online education platform based on XAPI Bayesian fuzzy rough set is established. Firstly, stratified, then its correlation with effect are analysed, indicators determined. Finally, The experimental results show that accuracy always above 93%, high; missing between 1% 4%, low; maximum improvement 14.8%, students' high, which verifies...

A variational model is proposed for color image fusion and contrast enhancement simultaneously. It combines the geometry of images with coherence correlation constraints, perceptual regularity constraints into a framework. The gradient descent flow applied to minimize functional numerical scheme PDEs presented. compared wavelet-based approach visually quantitatively. Experimental results on multi-focus multi-spectral remote sensing are shown effectiveness verified. For images, can recover an...