In recent years, deep learning has made breakthroughs in the field of computer vision, single-stage detection algorithm represented by You Only Look Once (YOLO) achieved satisfying results SAR ship target detection. For multi-scale problem targets complex scenes, we proposed an improved YOLOv5 method using Convolutional Block Attention Module (CBAM) and Bidirectional Feature Pyramid Network (BiFPN). The CBAM module BiFPN are added so that it can fully learn feature information space channel...

Geospatial object detection from high spatial resolution (HSR) remote sensing imagery is a heated and challenging problem in the field of automatic image interpretation. Despite convolutional neural networks (CNNs) having facilitated development this domain, computation efficiency under real-time application accurate positioning on relatively small objects HSR images are two noticeable obstacles which have largely restricted performance methods. To tackle above issues, we first introduce...

For deployment on an embedded processor for distracted driver classification, the model should satisfy demand both high accuracy, real-time inference, and limited storage resources. Conventional deep CNN models such as VGG, ResNet, DenseNet, often aim making their heavy system with memory space computing In contrast, lightweight are greatly compressed but at a significant sacrifice of accuracy. To bridge this gap, we propose instance-specific multi-teacher knowledge distillation (IsMt-KD) to...

Proteins play a pivotal role in living organisms, yet understanding their functions presents significant challenges, including the limited flexibility of classification-based methods, inability to effectively leverage spatial structural information, and lack systematic evaluation metrics for protein Q&A systems. To address these limitations, we propose Prot2Chat, novel framework that integrates multimodal representations with natural language through unified module, enabling large model...

Safety helmet-wearing detection is an essential part of the intelligent monitoring system. To improve speed and accuracy detection, especially small targets occluded objects, it presents a novel efficient detector model. The underlying core algorithm this model adopts YOLOv5 (You Only Look Once version 5) network with best comprehensive performance. It improved by adding attention mechanism, CIoU (Complete Intersection Over Union) Loss function, Mish activation function. First, applies...

As a basic task in the field of computer vision, target detection has been concerned by many researchers. The performance method is also directly related to research advanced semantic fields. Current general methods are not effective small detection, so this paper studies problem and proposes based on deep learning with considerate feature effectively expanded sample size. Firstly, according characteristics convolutional neural network, we improve current network architecture which performs...

Abstract Edge‐device‐based object detection is crucial in many real‐world applications, such as self‐driving cars, ADAS, driver behavior analysis. Although deep learning (DL) has become the de‐facto approach for detection, limited computing resources of embedded devices and large model size current DL‐based methods increase difficulty real‐time on edge devices. To overcome these difficulties, this work a novel YOLOv4‐dense proposed to detect objects an accurate, fast manner, which built top...

With the rapid development of remote sensing technology, application convolutional neural networks in object detection has become very widespread, and some multimodal feature fusion have also been proposed recent years. However, these methods generally do not consider long-tailed problem that is widely present images, which limits further improvement model performance. To solve this problem, we propose a novel method for can effectively fuse complementary information visible light infrared...

Automatic ship detection in SAR imagery plays an indispensable role the surveillance of maritime activity. With spring deep neural network and rapid development imaging technique, based on convolutional networks has attracted increasing attention remote sensing interpretation. However, ships variant sizes (multi-scale) complicated background result essential challenge using learning methods. The initial motivation this paper is to design a structure that can achieve better accuracy as well...

The accurate and timely abnormal object detection is of crucial importance for the safe operation power grid. It rather difficult, however, to completely manually recognize such objects based on uploaded pictures in cloud server. To meet demand accuracy timeliness, this paper proposes combine cloud/edge fusion framework deep learning techniques detection. Specifically, we first train model by using YOLOv4 server, then apply trained detect whether there an each captured picture edge servers....

3-D object detection from mobile phones in Device-to-Device (D2D) system provides a new smart payment tool for the next generation of fintech, which is more flexible and efficient than traditional barcode. In this article, we propose monocular method based on depth-guided local convolution. The combines information RGB image mode depth by using convolution kernel through works single locally. According to multiscale input information, adaptively adjusted capture target objects different...

Federated learning (FL) is a promising framework for privacy-preserving collaborative learning, where model training tasks are distributed to clients and only the updates need be collected at server. However, when being deployed mobile edge networks, may have unpredictable availability drop out of process, which hinders convergence FL. This paper tackles such critical challenge. Specifically, we first investigate classical FedAvg algorithm with arbitrary client dropouts. We find that common...

A large number of forensics research focus on operation detection to reveal the evidence forgery action in digital image. In early works, analyst firstly model probability distribution single operation, and design forensic tools based feature extraction machine learning classifier. With increasing dimension facing multiple operations scenario, physical meaning gradually become ambiguous. Especially, since deep algorithm was used research, automatic selection making decision with high...

For the problems of large size and high computation deep convolutional neural network (CNN) models, which make it difficult to achieve real-time detection on resource-limited mobile or embedded devices, existing lightweight models are not sufficient for small targets distribution side, a model compression algorithm that combines multi-scale target recognition PM-YOLO (Prune-MobileNetv3-YOLOv5s) is proposed efficient detection. Based YOLOv5s model, feature information performed at different...

Accurate segmentation and classification of nuclei in histology images is critical but challenging due to heterogeneity, staining variations, tissue complexity. Existing methods often struggle with limited dataset variability, patches extracted from similar whole slide (WSI), making models prone falling into local optima. Here we propose a new framework address this limitation enable robust nuclear analysis. Our method leverages dual-level ensemble modeling overcome issues stemming...

Synthetic aperture radar (SAR) is an active microwave imaging sensor, which can provide images in all-weather and all-day conditions. The various scales irregular distribution of different ships SAR images, a heated challenging problem. As basic component the object detection frameworks, Feature Pyramid Networks (FPNs) improve feature representations for detecting objects at scales. However, FPN adopts same convolution operation layers, does not consider differences between levels. In this...

In traditional multi-target tracking algorithm, only target kinematic information has been used for data association. A new association algorithm is presented in this paper-Feature Aided Tracking (FAT) which based on the Generalized Probability Data Association (GPDA) algorithm. FAT combines feature with a probabilistic way, preferably resolves closely spaced targets dense clutter environment. This idea demonstrated via an example where ID range profile measurement incorporated into...

3D object detection is a challenging problem in computer vision. In this paper, Point-Voxel Feature Encoder (PVFE) proposed to obtain the spatial context information between points and voxels for point clouds. Firstly, cloud converted into voxel grids avoid loss of structure projecting. Then, voxel-wise features are learned through PVFE, which consisting stacked fully connected layer. Last, processed by Sparse Convolutional Layers as input RPN performs category prediction bounding box...

Car detection in the wild suffers from problem of high occlusion and complicated scene disturbance. In this paper, we propose a novel Context Combined Cascaded Region Proposal Network (C3RPN) with dual loss function which addresses these problems at three dimensions: (i) cascaded two-stage region proposal structure that removes limitation fixed anchor boxes, (ii) Smooth L1 Intersection over Union (IoU) are adopted to optimize two stages respectively for better convergence finer regression,...

Recent top-performing object detectors usually depend on a two-stage approach, which benefits from its region proposal and refining practice but suffers low detection speed. By contrast, one-stage approaches have the advantage of high efficiency while sacrifice their accuracies to some extent. In this paper, we propose novel single-shot network inherits merits both. Motivated by idea semantic enrichment convolutional features within typical deep detector, two modules: 1) modeling...