The diagnosis of blood-related diseases involves the identification and characterization a patient's blood sample. As such, automated methods for detecting classifying types cells have important medical applications in this field. Although deep convolutional neural network (CNN) traditional machine learning shown good results classification cell images, they are unable to fully exploit long-term dependence relationship between certain key features images image labels. To resolve problem, we...

In order to resist the adverse effect of viewpoint variations, we design quadruple directional deep learning networks extract features (QD-DLF) vehicle images for improving re-identification performance. The are similar overall architecture, including same basic architecture but different feature pooling layers. Specifically, that is a shortly and densely connected convolutional neural network utilized maps an input square image in first stage. Then, utilize layers, i.e., horizontal average...

Fog computing, as the supplement of cloud can provide low-latency services between mobile users and cloud. However, fog devices may encounter security challenges a result nodes being close to end having limited computing ability. Traditional network attacks destroy system nodes. Intrusion detection (IDS) is proactive protection technology be used in environment. Although IDS tradition has been well investigated, unfortunately directly using them environment inappropriate. produce massive...

Low-resolution medical images can hamper diagnosis seriously, especially in the analysis of retina and specifically for detection macula fovea. Therefore, improving quality speeding up their reconstruction is particularly important expert diagnosis. To deal with this engineering problem, our paper presents a fast image super-resolution (FMISR) method whereby three hidden layers to complete feature extraction as same super resolution convolution neural network. It that well-designed deep...

Mobile cloud computing (MCC) integrates (CC) into mobile networks, prolonging the battery life of users (MUs). However, this mode may cause significant execution delay. To address delay issue, a new known as edge (MEC) has been proposed. MEC provides and storage service for network, which enables MUs to execute applications efficiently meet requirements. In paper, we present comprehensive survey research from perspective adoption provision. We first describe overview MEC, including...

Mobile edge computing is becoming a promising architecture to overcome the resource limitation of mobile devices and bandwidth bottleneck core networks in cloud computing. Although offloading applications can extend performance for devices, it may also lead greater processing latency. Usually, users have pay cloudlet or they used. In this paper, we bring thorough study on energy consumption, time cost using workflow Based theoretical analysis, multi-objective optimization model established....

Person re-identification (Re-ID) aims to match person images captured from two non-overlapping cameras. In this paper, a deep hybrid similarity learning (DHSL) method for Re-ID based on convolution neural network (CNN) is proposed. our approach, light CNN feature pair the input image simultaneously extracted. Then, both elementwise absolute difference and multiplication of are calculated. Finally, function designed measure between pair, which realized by group weight coefficients project...

The classification of benign and malignant based on ultrasound images is great value because breast cancer an enormous threat to women’s health worldwide. Although both texture morphological features are crucial representations tumor images, their straightforward combination brings little effect for improving the since high-dimensional too aggressive so that drown out low-dimensional features. For that, efficient feature combing method proposed improve malignant. Firstly, (i.e., local binary...

In this paper, a joint feature and similarity deep learning (JFSDL) method for vehicle reidentification is proposed. The proposed JFSDL applies siamese network to extract features an input image pair simultaneously. learned under the identification verification supervision. supervision realized by linearly combining two softmax functions one hybrid function. Moreover, based on function, score between also obtained simultaneously projecting element-wise absolute difference multiplication of...

In this paper, we propose a shortly and densely connected convolutional neural network (SDC-CNN) for vehicle re-identification. The proposed SDC-CNN mainly consists of short dense units (SDUs), necessary pooling normalization layers. main contribution lies at the design connection mechanism, which would effectively improve feature learning ability. Specifically, in each SDU contains list layers layer is same appropriate channels. Consequently, number connections input channel are limited...

We designed a gangue sorting system,and built convolutional neural network model based on AlexNet. Data enhancement and transfer learning are used to solve the problem which convolution has insufficient training data in stage. An object detection region clipping algorithm is proposed adjust image optimum size. Compared with traditional SVM algorithm, this higher recognition rate for coal gangue, provides important reference identification separation of gangue.

In recent years, deep learning (DL) has been successfully and widely applied in the person reidentification (Re-ID). However, DL-based Re-ID methods face a bottleneck that scales of most existing databases are not large enough for training very models. To address this problem, body symmetry part-locality-guided direct nonparametric feature enhancement (DNDFE) method is proposed article. Based on observation part locality two important appearance properties inherited upright walking persons,...

Fabric image retrieval aims to match a probe fabric from gallery set, playing great role in the textile industry. However, there is lack of one comprehensive benchmark for retrieval. For this, large proposed this paper. Firstly, dataset namely 1.0 proposed, which contains 46,656 images 972 subjects. Each subject collected with multiple various angles and both front back sides. Secondly, evaluation protocol designed evaluate accuracy, uses cumulative characteristic (CMC) curve mean average...

Object reidentification with the goal of matching pedestrian or vehicle images captured from different camera viewpoints is considerable significance to public security. Quadruple directional deep learning features (QD-DLFs) can comprehensively describe object images. However, correlation among QD-DLFs an unavoidable problem, since are learned quadruple independent networks (QIDDNs) driven same training data, and each network holds basic feature architecture (BDFLA). The harmful...

Vehicle re-identification matching vehicles captured by different cameras has great potential in the field of public security. However, recent vehicle approaches exploit complex networks, causing large computations their testing phases. In this paper, we propose a behavior difference learning (MBDL) method to compress models for saving computations. order represent evolution across two layers deep network, (MBD) matrix is designed. Then, our MBDL minimizes L1 loss function among MBD matrixes...

Human key point detection has essential application prospects in human and computer interaction disease prediction. However, the current scenario, there is still some room for improvement accuracy of when facing complex scenes points are missing. To address above situation this paper proposes an improved YOLOv8-SW model based on YOLOV8 model, adding SA attention mechanism to improve model's ability extract channel spatial features, replacing IOU with WIOU, increasing proportion weight...

Agriculture is vital to human survival and remains one of the main driving forces several economies in world, more so developing economies. an important industry China. With increase agricultural demand, it urgent reduce costs while maximizing production. As there are few traditional algorithms for enoki mushroom detection, this paper proposed a automatic caps classification algorithm, built convolutional neural network model based on LeNet. The existing preprocessing approaches models...