Average Precision Rate Research Articles

A CNN-based posture change detection for lactating sow in untrimmed depth videos

The dangerous body movements of lactating sows are mainly reflected in the frequency and duration of posture changes. To avoid the disturbances from the light environment and illumination variations caused by heat lamp and day-night cycle, depth videos were acquired for posture change detection. In this paper, a CNN-based method was developed to automatically detect sow posture change in untrimmed depth videos. Firstly, clip-level Detect and Classify Posture R-CNN (DCP R-CNN) was proposed for sow detection and posture classification. The DCP R-CNN decoupled the detection and classification task into two branches: detection branch and classification branch so that the detection branch could focus on the task of sow localization. Secondly, a method was proposed to temporally localize the possible posture changing segments from coarse to fine. Finally, a two-stream posture change classification network, which used two CNNs to extract the complementary features from DMM and the first and the last frames, was introduced to classify the possible posture changing segments into change or non-change. At the scene of commercial piggery, the videos were acquired at 5 frames per second (fps) by installing a Microsoft Kinect v2.0 sensor above the pen. The test set consisted of two 24-hour untrimmed depth videos and six 40-minute untrimmed depth videos. The test results showed that our method could detect the posture change correctly. The success rate of DCP R-CNN was 96.38% when the IoU threshold was 0.7, and the average precision and recall rate of 4 types of posture were 97.72% and 95.23%, respectively. The possible posture change was detected with a recall rate of 92.46% and the posture change was classified with an accuracy of 97.71%. The precision of posture change detection was 88.75%, the recall rate was 90.87%. The detection speed of our method was 5.71 fps, which was faster than the video frame rate (5 fps). Our study demonstrated the feasibility of 24-hour detection method for sow posture change.

Facial expression analysis using local directional stigma mean patterns and convolutional neural networks

This paper represents automatic facial expression analysis method named Local Directional Stigma Mean Patterns (LDSMP) for automatic facial expression analysis and image retrieval using content based facial expression image retrieval and CNN. The traditional local patterns such as Local Binary Patterns (LBP) and Local Ternary Patterns (LTP) are applied for face recognition and expression analysis, calculated using relationship between the center pixel and neighboring pixels. The proposed method calculates the eight directional difference values then divided into the three ranges based on threshold values. Thus, the values are substituted with basic three positive values (+3, +2, +1) and three negative values (-3, -2, -1) to get more sensitive information from an image rather than aforementioned methods. The threshold can be select either static which is selected by user or dynamic is evaluated from image itself and supports to improve the efficiency. The performance of the proposed method is further improved by giving this patterns as input to the Convolutional Neural Networks (CNN) and compared with the existing methods LBP, LTP, and Directional Binary Code (DBC) in terms of Average Precision (AP), Average Recall (AR), and Average Retrieval Rate (ARR) using standard databases COREL 10K (DB1) and JAFFE (The Japanese Female Facial Expression) (DB2) and Extended Cohn-Kanade (CK +) (DB3) dataset.

A task recommendation framework for heterogeneous mobile crowdsensing

Aiming at the problems of low data quality and high incentive costs caused by the low enthusiasm of participants in mobile crowd sensing, a new task recommendation framework is proposed in this paper. First, the participants' historical behaviors are analyzed, assuming that user behaviors can be quantified as the user's willingness to participate, and the cosine similarity theorem is used to calculate the similarity between participants, thereby constructing a user-hybrid model. Secondly, probabilistic matrix factorization is developed to predict the willingness of participants, and a ranking model is obtained through learn-to-rank algorithm. Finally, a task recommendation list is generated according to the ranking model, which serves as the target participant's preferred task list for sensing task recommendation. The experiment in this paper is carried out on the MATLAB platform based on two real check-in datasets, Gowalla and Brightkite. The results show that the average allocation precision rate can reach 96%, and the sensing user participation rate is about 97%. Meantime, the user's mobile cost is reduced, and the overall goal of maximizing accuracy and minimizing perceived cost is achieved.

Detection of partially occluded objects – A comparative analysis based on Haar Classifier and K-Means Clustering

This paper addresses the problem of detecting partially occluded objects from 2D images. The detection of partially occluded objects is performed and compared using feature-based training and color-based object segmentation. The occluded objects are very difficult to be detected based only on their features since, all the essential features may not be visible to the learned model due to occlusion. Haar cascade classifier has been utilised for feature-based training and the k-means clustering is utilized for color-based tracking. Various input images are provided Haar classifier as well as the K-means clustering to detect the objects in the 2D images and the subsequent results are compared and analysed. For segmenting the 2D objects using k-means clustering, the average recall and average precision varies from 0.70 to 0.98. The variation is based upon the veracity of the occluded objects. The average precision rate for detecting the occluded 2D objects through the developed method is between 0.24 and 0.60. And it is noted that the average recall for the respective detection lies between 0.25 and 0.70.

Bubble recognizing and tracking in a plate heat exchanger by using image processing and convolutional neural network

Water and air are usually employed as a heat exchange medium in cold channels of a plate heat exchanger (PHE), while air, in the state of bubbles in water, has an apparent impact on PHE performance, such as heat exchanging efficiency, flow resistance, etc. However, individual bubble behavior, such as bubble rupturing, merging, colliding, etc., are difficult to detect due to the flow complexity in PHE. Aiming at the problem of exploring individual bubble behavior visually, this study proposes a new method to recognize and track the bubbles in PHE based on a visualization bench for the cold channel of a dimple-type embossing PHE. Firstly, convolutional neural network (CNN) and improved three-frame difference (ITFD) method are used to detect and attain the position and state of the bubble flow in the transparent passage of the PHE from captured videos. Then, the intersection-over-union (IOU) screening algorithm is adopted to optimize the results. Finally, the bubble positions and velocities are calculated. Furthermore, dimensionless parameters such as the local Reynolds number, Weber number, and Froude number are also obtained. The results show that the proposed method could precisely recognize and track individual bubble's spatiotemporal behavior, such as rupturing, merging, and colliding. In the presence of a large number of dense bubbles in the channel of a PHE, this method can achieve an average precision rate of over 94 %, a recall rate of over 87 %, and F1 score of 0.91.

Sensor Fusion Basketball Shooting Posture Recognition System Based on CNN

In recent years, with the development of wearable sensor devices, research on sports monitoring using inertial measurement units has received increasing attention; however, a specific system for identifying various basketball shooting postures does not exist thus far. In this study, we designed a sensor fusion basketball shooting posture recognition system based on convolutional neural networks. The system, using the sensor fusion framework, collected the basketball shooting posture data of the players’ main force hand and main force foot for sensor fusion and used a deep learning model based on convolutional neural networks for recognition. We collected 12,177 sensor fusion basketball shooting posture data entries of 13 Chinese adult male subjects aged 18–40 years and with at least 2 years of basketball experience without professional training. We then trained and tested the shooting posture data using the classic visual geometry group network 16 deep learning model. The intratest achieved a 98.6% average recall rate, 98.6% average precision rate, and 98.6% accuracy rate. The intertest achieved an average recall rate of 89.8%, an average precision rate of 91.1%, and an accuracy rate of 89.9%.

Apple Fruit Recognition Based on a Deep Learning Algorithm Using an Improved Lightweight Network

HIGHLIGHTSA deep learning algorithm with an improved lightweight network was used to identify apple fruit.Multiscale pooling was used to reduce the image size and enrich the features.Compound scaling was used to scale the basic network.The optimal compound scaling coefficient for apple identification was obtained with the WOA algorithm.The proposed method achieved an average recognition precision rate of 94.43% and a speed of 0.051s.ABSTRACT. Accurate fruit identification is the basis for automating the operation of orchard production. To better apply the identification model in mobile devices so that venue becomes a less restrictive factor for application, this study proposes an apple fruit identification method based on an improved lightweight network named “MobileNetV3-Small.” The whale optimization algorithm was introduced to improve the model by obtaining an optimal compound-scaling coefficient for the MobileNetV3-Small network. A multiscale pooling approach was used for fruit recognition, comprising operations such as lossless scaling and feature extraction on sample images. The obtained images were then inputted into the model for recognition and classification. The experimental process was conducted on an apple data set. The test results show that the multiclass average precision of apple recognition using this model was 94.43% and the running time of recognition was 0.051 s per image. Both indicators outperformed the control network models of “MobileNetV3-Small,” ResNet-50, and VGG-19. This model is 14.63% more accurate and 1.95 times quicker on average in identification than the next best model. These findings indicate that the method can realize high-efficiency and high-precision recognition of apples with high stability and portability, which lays a good foundation for the mechanization of repetitive operations such as orchard yield estimation, fruit labeling, and fruit picking. Keywords: Apple recognition, Compound scaling, Deep learning algorithm, Lightweight network, Yield estimation.

Limited sliding network: Fine‐grained target detection on electrical infrastructure for power transmission line surveillance

SummaryBecause of its small size, low local contrast, and much interference, the field image of fine‐grained equipment taken from power transmission line surveillance is hard to be sustained by the traditional small target detection technique, which requires the manual extraction of features, making it difficult to accurately detect micro‐fine‐grained equipment. The deep learning‐based algorithms have prospective application but require abundant data to guarantee performance and tackle the problem of foreground–background imbalance. This paper develops an effective pipeline, i.e., limited sliding network (LSNet), to detect the small and fine‐grained defects on equipment in power transmission line infrastructure. The model firstly performs the regional analysis on the entire image to determine the potential target locations. The feature extraction and classification on the potential location image blocks are further performed by the VGG‐style model for the dense target locations, and the nonmaximum suppression method is finally applied to locate the target. On the other hand, a specific training method is also developed to better deal with a wide range imbalances of positive and negative samples. The proposed method achieves the detection mean average precision (mAP) rate of 98.66% on the real datasets, while limiting the computational overhead of hardware.

Deep Fuzzy Multi-Object Categorization in Scene

Object Categorization is the process of, identifying and labelling the various distinct Classes (Categories), in the given input image. The Deep Fuzzy Multi-Object Categorization (DFMOC) model, combines the learning capability of Convolution Neural Networks (CNN) and the uncertainty-managing ability of Fuzzy system, for carrying out the categorization task. This work starts with Background Elimination process for ensuring the image clarity, followed by Fuzzification and Fuzzy Entropy computation. Simple fuzzy sets are to be framed, by employing Fuzzy C-Means (FCM) algorithm, for fuzzification of the input image. Thresholding Block is incorporated, for determining the clusters . The Fuzzy Entropy Computation (FEC) is done, to minimize the Fuzziness rate of the acquired input and consequently, the layers of CNN are trained in accordance with that. Caltech-101 Dataset is been utilized for analysis. Average Precision Rate of Categorization (APRC), along with other metrics namely Time taken and Error Rate, shows that DFMOC model performs better than other models.

Stacked-autoencoder-based model for COVID-19 diagnosis on CT images.

With the outbreak of COVID-19, medical imaging such as computed tomography (CT) based diagnosis is proved to be an effective way to fight against the rapid spread of the virus. Therefore, it is important to study computerized models for infectious detection based on CT imaging. New deep learning-based approaches are developed for CT assisted diagnosis of COVID-19. However, most of the current studies are based on a small size dataset of COVID-19 CT images as there are less publicly available datasets for patient privacy reasons. As a result, the performance of deep learning-based detection models needs to be improved based on a small size dataset. In this paper, a stacked autoencoder detector model is proposed to greatly improve the performance of the detection models such as precision rate and recall rate. Firstly, four autoencoders are constructed as the first four layers of the whole stacked autoencoder detector model being developed to extract better features of CT images. Secondly, the four autoencoders are cascaded together and connected to the dense layer and the softmax classifier to constitute the model. Finally, a new classification loss function is constructed by superimposing reconstruction loss to enhance the detection accuracy of the model. The experiment results show that our model is performed well on a small size COVID-2019 CT image dataset. Our model achieves the average accuracy, precision, recall, and F1-score rate of 94.7%, 96.54%, 94.1%, and 94.8%, respectively. The results reflect the ability of our model in discriminating COVID-19 images which might help radiologists in the diagnosis of suspected COVID-19 patients.

Identifying click-requests for the network-side through traffic behavior

With the rapid development of web-based applications, clicking on hyperlinks has become a general means for accessing various network services. Understanding the visiting behavior of web users not only helps improve the personalized service quality and user experience, but also plays an important role in network management and early threat detection. Click-stream identification is a fundamental issue for user behavior analysis. However, most existing approaches are designed for non-encrypted HTTP requests and only focus on server-side scenarios, which makes them inapplicable to the increasingly popular HTTPS and network-side management. In this work, we propose an encryption-independent scheme from a network-side perspective that adopts the web traffic collected at the network boundary to identify the HTTP(S) requests generated by the click actions of web users. The proposed scheme employs hidden Markov models (HMMs) to describe the time-varying behavior of click and non-click web traffic. A deep neural network (DNN) is integrated into the HMMs to capture the context of web traffic, which eliminates the limitations caused by the independence hypothesis of the traditional HMMs. Finally, a DNN-based rear classifier is proposed to determine the type of HTTP(S) requests according to the fitting degree between the HTTP(S) requests and the HMM-based behavior models. We derive the algorithms for model learning and click identification. Experiments are conducted to validate the proposed approach. Performance-related issues and comparisons are discussed. Results show that both the average precision and recall rate of the proposed approach exceed 92%, which is better than most existing benchmark methods in terms of performance and stability.

Automatic tooth roots segmentation of cone beam computed tomography image sequences using U-net and RNN.

Automatic segmentation of individual tooth root is a key technology for the reconstruction of the three-dimensional dental model from Cone Beam Computed Tomography (CBCT) images, which is of great significance for the orthodontic, implant and other dental diagnosis and treatment planning. Currently, tooth root segmentation is mainly done manually because of the similar gray of the tooth root and the alveolar bone from CBCT images. This study aims to explore the automatic tooth root segmentation algorithm of CBCT axial image sequence based on deep learning. We proposed a new automatic tooth root segmentation method based on the deep learning U-net with AGs. Since CBCT sequence has a strong correlation between adjacent slices, a Recurrent neural network (RNN) was applied to extract the intra-slice and inter-slice contexts. To develop and test this new method for automatic segmentation of tooth roots using CBCT images, 24 sets of CBCT sequences containing 1160 images and 5 sets of CBCT sequences containing 361 images were used to train and test the network, respectively. Applying to the testing dataset, the segmentation accuracy measured by the intersection over union (IOU), dice similarity coefficient (DICE), average precision rate (APR), average recall rate (ARR), and average symmetrical surface distance (ASSD) are 0.914, 0.955, 95.8% , 95.3% , 0.145 mm, respectively. The study demonstrates that the new method combining attention U-net with RNN yields the promising results of automatic tooth roots segmentation, which has potential to help improve the segmentation efficiency and accuracy in future clinical practice.

Hyperspectral remote sensing image classification based on tighter random projection with minimal intra-class variance algorithm

Aiming at solving the problem of image size limiting in the traditional Random Projection (RP) algorithm, a novel Tighter Random Projection (TRP), which combines the scheme with Minimal Intra-class Variance (TRP-MIV) for hyperspectral remote sensing image classification is proposed. First, a new tighter dimensional boundary for expanding image size with the TRP-MIV matrix selected by multiple sampling for improving the class separability is defined to reduce dimension. Then the proposed algorithm is implemented, which integrates TRP-MIV for dimensionality reduction and Minimum Distance (MD) classifier for image classification. Finally, the image size and dimensionality reduction are evaluated by the number of spectral pixels under different theorems, and the spectral difference before and after dimensionality reduction, respectively. Classification performance is evaluated by kappa coefficient, Overall Accuracy (OA), Average Accuracy (AA), Average Precision Rate (APR) and running time. Classification results are obtained from the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) scanner and the Reflective Optics System Imaging Spectrometer (ROSIS) scanner, which indicate that the proposed algorithm is efficient and promising.

Deep learning for the determination of myometrial invasion depth and automatic lesion identification in endometrial cancer MR imaging: a preliminary study in a single institution.

To determine the diagnostic performance of a deep learning (DL) model in evaluating myometrial invasion (MI) depth on T2-weighted imaging (T2WI)-based endometrial cancer (EC) MR imaging (ECM). We retrospectively enrolled 530 patients with pathologically proven EC at our institution between January 1, 2013, and December 31, 2017. All imaging data were reviewed on picture archiving and communication systems (PACS) server. Both sagittal and coronal T2WI-based MR images were used for lesion area determination. All MR images were divided into two groups: deep (more than 50%) and shallow (less than 50%) MI based on their pathological diagnosis. We trained a detection model based on YOLOv3 algorithm to locate the lesion area on ECM. Then, the detected regions were fed into a classification model based on DL network to identify MI depth automatically. In the testing dataset, the trained model detected lesion regions with an average precision rate of 77.14% and 86.67% in both sagittal and coronal images, respectively. The classification model yielded an accuracy of 84.78%, a sensitivity of 66.67%, a specificity of 87.50%, a positive predictive value of 44.44%, and a negative predictive value of 94.59% in determining deep MI. The radiologists and trained network model together yielded an accuracy of 86.2%, a sensitivity of 77.8%, a specificity of 87.5%, a positive predictive value of 48.3%, and a negative predictive value of 96.3%. In this study, the DL network model derived from MR imaging provided a competitive, time-efficient diagnostic performance in MI depth identification. • The models established with the deep learning method could help improve the diagnostic confidence and performance of MI identification based on endometrial cancer MR imaging. • The models enabled the classification of endometrial cancer MR images to the two categories with a sensitivity of 0.67, a specificity of 0.88, and an accuracy of 0.85. • Using the detected lesion region to evaluate myometrial invasion depth could remove redundant information in the image and provide more effective features.

A Data-Flow Oriented Deep Ensemble Learning Method for Real-Time Surface Defect Inspection

Real-time surface defect inspection plays an important role in the quality control of automated production. For the surface defect inspection, flowing data are a common data form in the automated pipeline production. Although flowing data provide rich information for surface defect inspection, there are still a lot of dynamic distribution challenges caused by the flowing data, such as the domain shift phenomenon and imbalanced training data. However, many existing industrial inspection solutions still use static strategies. To determine the dynamic distribution influence on the data flow domain, this article proposes a new deep ensemble learning method with domain fluctuation adaptation. Specifically, a new distribution discrepancy identifier based on estimation of the data set distribution and data characteristic is proposed. It utilizes advantages of both the deep convolutional neural network (CNN) and the shallow feature-based learning method to achieve higher robustness and fine-grained detection in streaming data scenes. In order to validate the proposed method, an inspection bench test system, as a part of a real industrial surface mount technology production line, is designed and fabricated. The proposed inference model is successfully applied to an embedded terminal with a hybrid and heterogeneous computing architecture. At last, the method is validated on the data collected from the manufacturer. The result suggests that the proposed method possesses a competitive mean average precision (mAP) rate with good adaptation and robustness in industrial streaming data scenes.

Learning image representation from image reconstruction for a content-based medical image retrieval

In this paper, we propose a novel approach of feature learning through image reconstruction for content-based medical image retrieval. We propose an image reconstruction network to encode the input image into a set of features followed by the reconstruction of the input image from the encoded features. The robust reconstruction of the input image from encoded features shows that the encoded features can be used as an abstract version of an input image. Thus, we make use of these encoded features for medical image retrieval task. The performance of the proposed method has been analyzed with the help of three benchmark medical image databases. Average retrieval rate and average precision rate are used to evaluate the performance of proposed and existing state-of-the-art methods for medical image retrieval task. Experimental analysis shows that the proposed approach for image retrieval outperforms the other existing methods.

Video-Based Detection Infrastructure Enhancement for Automated Ship Recognition and Behavior Analysis

Video-based detection infrastructure is crucial for promoting connected and autonomous shipping (CAS) development, which provides critical on-site traffic data for maritime participants. Ship behavior analysis, one of the fundamental tasks for fulfilling smart video-based detection infrastructure, has become an active topic in the CAS community. Previous studies focused on ship behavior analysis by exploring spatial-temporal information from automatic identification system (AIS) data, and less attention was paid to maritime surveillance videos. To bridge the gap, we proposed an ensemble you only look once (YOLO) framework for ship behavior analysis. First, we employed the convolutional neural network in the YOLO model to extract multi-scaled ship features from the input ship images. Second, the proposed framework generated many bounding boxes (i.e., potential ship positions) based on the object confidence level. Third, we suppressed the background bounding box interferences, and determined ship detection results with intersection over union (IOU) criterion, and thus obtained ship positions in each ship image. Fourth, we analyzed spatial-temporal ship behavior in consecutive maritime images based on kinematic ship information. The experimental results have shown that ships are accurately detected (i.e., both of the average recall and precision rate were higher than 90%) and the historical ship behaviors are successfully recognized. The proposed framework can be adaptively deployed in the connected and autonomous vehicle detection system in the automated terminal for the purpose of exploring the coupled interactions between traffic flow variation and heterogeneous detection infrastructures, and thus enhance terminal traffic network capacity and safety.

Automatic detection of periodontal compromised teeth in digital panoramic radiographs using faster regional convolutional neural networks.

PurposePeriodontal disease causes tooth loss and is associated with cardiovascular diseases, diabetes, and rheumatoid arthritis. The present study proposes using a deep learning-based object detection method to identify periodontally compromised teeth on digital panoramic radiographs. A faster regional convolutional neural network (faster R-CNN) which is a state-of-the-art deep detection network, was adapted from the natural image domain using a small annotated clinical data- set.Materials and MethodsIn total, 100 digital panoramic radiographs of periodontally compromised patients were retrospectively collected from our hospital's information system and augmented. The periodontally compromised teeth found in each image were annotated by experts in periodontology to obtain the ground truth. The Keras library, which is written in Python, was used to train and test the model on a single NVidia 1080Ti GPU. The faster R-CNN model used a pretrained ResNet architecture.ResultsThe average precision rate of 0.81 demonstrated that there was a significant region of overlap between the predicted regions and the ground truth. The average recall rate of 0.80 showed that the periodontally compromised teeth regions generated by the detection method excluded healthiest teeth areas. In addition, the model achieved a sensitivity of 0.84, a specificity of 0.88 and an F-measure of 0.81.ConclusionThe faster R-CNN trained on a limited amount of labeled imaging data performed satisfactorily in detecting periodontally compromised teeth. The application of a faster R-CNN to assist in the detection of periodontally compromised teeth may reduce diagnostic effort by saving assessment time and allowing automated screening documentation.

Content Based Image Retrieval using Feature Extraction Technique

Now days the image processing can be used in various areas such as in Agriculture, in Health care system also for security purpose. In case of Crime investigation the image processing can be used to identify the particular suspect from an available dataset for that purpose an image retrieval technique is presented in this paper. For image retrieval number of techniques is available. In earlier days Block Truncation Coding is used but due its some disadvantage feature extraction method is used. Using DDBTC technique two features are derived. The first feature as Color Co-occurrence Features (CCF) obtained using color quantizes features such as Bit Pattern Feature (BPF) is derived from Bitmap image. The five different distance metrics are used to measure the similarity between two images. The simulated results shows proposed Technique can shows the better result in the form of Average Precision rate (APR) and Average Recall Rate (ARR) as compared to other techniques.

A universal cross language software similarity detector for open source software categorization

While there are novel approaches for detecting and categorizing similar software applications, previous research focused on detecting similarity in applications written in the same programming language and not on detecting similarity in applications written in different programming languages. Cross-language software similarity detection is inherently more challenging due to variations in language, application structures, support libraries used, and naming conventions. In this paper we propose a novel model, CroLSim, to detect similar software applications across different programming languages. We define a semantic relationship among cross-language libraries and API methods (both local and third party) using functional descriptions and a word-vector learning model. Our experiments show that CroLSim can successfully detect cross-language similar software applications, which outperforms all existing approaches (mean average precision rate of 0.65, confidence rate of 3.6, and 75% highly rated successful queries). Furthermore, we applied CroLSim to a source code repository to see whether our model can recommend cross-language source code fragments if queried directly with source code. From our experiments we found that CroLSim can recommend cross-language functional similar source code when source code is directly used as a query (average precision=0.28, recall=0.85, and F-Measure=0.40).