Proposed Detection Model Research Articles

A Magnified Adaptive Feature Pyramid Network for automatic microaneurysms detection

Diabetic retinopathy (DR), as an important complication of diabetes, is the primary cause of blindness in adults. Automatic DR detection poses a challenge which is crucial for early DR screening. Currently, the vast majority of DR is diagnosed through fundus images, where the microaneurysm (MA) has been widely used as the most distinguishable marker. Research works on automatic DR detection have traditionally utilized manually designed operators, while a few recent researchers have explored deep learning techniques for this topic. But due to issues such as the extremely small size of microaneurysms, low resolution of fundus pictures, and insufficient imaging depth, the DR detection problem is quite challenging and remains unsolved. To address these issues, this research proposes a new deep learning model (Magnified Adaptive Feature Pyramid Network, MAFP-Net) for DR detection, which conducts super-resolution on low quality fundus images and integrates an improved feature pyramid structure while utilizing a standard two-stage detection network as the backbone. Our proposed detection model needs no pre-segmented patches to train the CNN network. When tested on the E-ophtha-MA dataset, the sensitivity value of our method reached as high as 83.5% at false positives per image (FPI) of 8 and the F1 value achieved 0.676, exceeding all those of the state-of-the-art algorithms as well as the human performance of experienced physicians. Similar results were achieved on another public dataset of IDRiD.

Deep Learning Model for the Inspection of Coffee Bean Defects

The detection of coffee bean defects is the most crucial step prior to bean roasting. Existing defect detection methods used in the specialty coffee bean industry entail manual screening and sorting, require substantial human resources, and are not standardized. To solve these problems, this study developed a deep learning algorithm to detect defects in coffee beans. The results reveal that when the pooling layer was used to enhance features and reduce neural dimensionality, some of the coffee been features were lost or misclassified. Therefore, a novel dimensionality reduction method was adopted to increase the ability of feature extraction. The developed model also overcame the drawbacks of padding causing blurred image boundaries and the dead neurons causing impeding feature propagation. Images of eight types of coffee beans were used to train and test the proposed detection model. The proposed method was verified to reduce the bias when classifying defects in coffee beans. The detection accuracy rate of the proposed model was 95.2%. When the model was only used to detect the presence of defects, the accuracy rate increased to 100%. Thus, the proposed model is highly accurate in coffee bean defect detection in the classification of eight types of coffee beans.

Multi-Class Object Learning with Application to Fabric Defects Detection

Deep convolutional neural networks (CNNs) have shown great success in single-class fabric image detection. However, real-world fabric defect images generally contain several types of defects in one image. Accurately recognizing and classifying multi-class fabric defect images is still an unsolved issue due to the complexity of intersected defects, as well as difficulty in distinguishing small-size defects. To address these challenges, this study develops a methodology based on the deep learning feature pyramid networks (FPN) approach to detect multi-class fabric defects. To evaluate the proposed detection model, we built a unique multi-class fabric defects database (DHU-MO1000), where multi-class defect images are generated by industrial monitors from a textile factory. We used the dataset as the benchmark for multi-class defects detection training and testing the FPN. Furthermore, we conducted extensive experimental validations for various design choices. The experimental results show that the model outperformed existing multi-class object detection methods.

Fast3DS: A real-time full-convolutional malicious domain name detection system

Detecting malicious domain names generated by domain generation algorithms is critical for defending the network against sophisticated attacks. In the past decade, deep-learning-based detection schemes have proven to be the most effective. However, each of these schemes requires sufficient computation time, which makes it difficult for real-time online detection. In this paper, we propose a real-time malicious domain name detection system which is called fast3DS. The proposed system contains a lightweight full-convolutional detection model, as well as a detection pipeline that contains multiple efficient schemes for high-speed data acquisition, filtering, and inference. The proposed detection model uses a parallel depthwise convolutional architecture to replace the standard convolution layer, and a lightweight global average pooling connection architecture is proposed to replace the fully connected layer, which can effectively reduce the parameters and computation time. To compensate for the decrease in accuracy due to model lightweighting, a lightweight attention mechanism is proposed to improve the accuracy of model detection. The proposed detection pipeline employs some industry-leading network traffic processing architecture and deep learning inference acceleration architecture, which can fully utilize the CPU for processing and computing. The experimental results denote that the proposed detection scheme can achieve accuracy close to the state-of-the-art with significantly fewer parameters, and the system can substantially improve the processing capability compared to the conventional detection system.

Multiple Leak Detection in Water Distribution Networks Following Seismic Damage

Water pipe leaks due to seismic damage are more difficult to detect than bursts, and such leaks, if not repaired in a timely manner, can eventually reduce supply pressure and generate both pollutant penetration risks and economic losses. Therefore, leaks must be promptly identified, and damaged pipes must be replaced or repaired. Leak-detection using equipment in the field is accurate; however, it is a considerably labor-intensive process that necessitates expensive equipment. Therefore, indirect leak detection methods applicable before fieldwork are necessary. In this study, a computer-based, multiple-leak-detection model is developed. The proposed technique uses observational data, such as the pressure and flow rate, in conjunction with an optimization method and hydraulic analysis simulations, to improve detection efficiency (DE) for multiple leaks in the field. A novel approach is proposed, i.e., use of a cascade and iteration search algorithms to effectively detect multiple leaks (with the unknown locations, quantities, and sizes encountered in real-world situations) due to large-scale disasters, such as earthquakes. This method is verified through application to small block-scale water distribution networks (WDNs), and the DE is analyzed. The proposed detection model can be used for efficient leak detection and the repair of WDNs following earthquakes.

Detection of Non-Technical Losses in Electric Distribution Network by Applying Machine Learning and Feature Engineering

Non-technical losses (NTL), which occur up to 40% of the total electric transmission and distribution power, create many challenges worldwide. These losses have a severe impact on distribution utilities and adversely affect the performance of electrical distribution networks. Furthermore, the depreciation of these NTL reduces the requirement of new power plants to fulfill the demand-supply gap. Hence, NTL is an emerging research area for electrical engineers. This paper proposed a model for the detection of non-technical losses based on machine learning and feature engineering. Experimental results check the performance of the proposed model. These results clearly show that this proposed detection model has better accuracy, precision, recall, F1 score, and AUC score than other existing approaches.

Development of Optimal Feature Selection and Deep Learning Toward Hungry Stomach Detection Using Audio Signals

The process of transmitting signals to the body regarding the hungry stomach is referred to as the migrating motor complex (MMC) process. The intestines and stomach are considered for sensing the unavailability of food in the body. Hence, the receptors present in the stomach wall generate the electrical activity waves and activate the hunger. In general, audio signal processing algorithms include signal analysis, property extraction, and behavior prediction, identifying the pattern available in the signal, and predicting how a specific signal is correlated to various identical signals. The major challenge here is to consider the audio signals that are produced from the stomach for identifying the growling sound that well describes the hungry. The main intent of this paper is to develop an intelligent model using the deep learning concept for recognizing the hungry stomach by using the synthetically collected audio signals through mobile phones. This makes society reaching the hungry stomach by way of intelligent technology. The proposed detection model covers different phases for automated hungry stomach detection. The data acquisition is performed by gathering information using mobile phones. Further, the pre-processing of the signals is done by the median filtering and the smoothening methods. In order to perform the proper classification, the spectral features like spectral centroid, spectral roll-off, spectral skewness, spectral kurtosis, spectral slope, spectral crest factor, and spectral flux, and cepstral domain features like mel-frequency cepstral coefficients (MFCCs), linear prediction cepstral coefficients (LPCCs), Perceptual linear prediction (PLP) cepstral coefficients, Greenwood function cepstral coefficients (GFCC), and gammatone cepstral coefficients (GTCCs) are extracted. Further, the optimal feature selection is done by the improved meta-heuristic algorithm called best and worst position updated deer hunting optimization algorithm (BWP-DHOA). An improved deep learning model called optimized recurrent neural network (RNN) is used for classifying the optimal features of the audio signal into growling sound and burp sound. Finally, the performance comparison over the existing models proves the efficiency and reliability of the proposed model.

An Improved SQL Injection Attack Detection Model Using Machine Learning Techniques

SQL Injection Attack (SQLIA) is a common cyberattack that target web application database. With the ever increasing and varying techniques to exploit web application SQLIA vulnerabilities, there is no a comprehensive method that can solve this kind of attacks. Therefore, these various of attack techniques required to establish many methods against in order to mitigate its threats. However, most of these methods have not yet been evaluated, where it is still just theories and require to implement and measure its performance and set its limitation. Moreover, most of the existing SQL injection countermeasures either used syntax-based detection methods or a list of predefined rules to detect the SQL injection, which is vulnerable in advance and sophisticated type of attacks because attackers create new ways to evade the detection utilizing their pre-knowledge. Although semantic-based features can improve the detection, up to our knowledge, no studies focused on extracting the semantic features from SQL stamens. This paper, investigates a designed model that can improve the efficacy of the SQL injection attack detection using machine learning techniques by extracting the semantic features that can effectively indicate the SQL injection attack. Also, a tenfold approach will be used to evaluate and validate the proposed detection model.

Detail texture detection based on Yolov4‐tiny combined with attention mechanism and bicubic interpolation

Aero-engine blades crack detection is one of the important tasks in daily ground maintenance, crack is a kind of texture feature, due to the random distribution, irregular shape and vague characteristics, which is still a challenging task to realize automatic detection in working environment. A detection model based on the Yolov4-tiny is proposed that is universal and focuses more on the characteristics of cracks, and it is implemented in embedded device. First, in order to distinguish the cracks and noises, an improved attention module is introduced into the backbone of Yolov4-tiny to enhance the model's capability to focus on crack areas; second, in order to improve the effect of multi-scale feature fusion, the bicubic interpolation is implemented in upsampling module; finally, in order to solve the redundant detection results of bounding-boxes in crack areas, the optimized non-maximum suppression method is proposed to make the detection results better corresponding to the groundTruth. The robustness of proposed detection model was demonstrated by evaluating varying lighting and noise images. The average precision on integrated datasets is 81.6%, which outperforms the original Yolov4-tiny by an increase of 12.3%.

Integration of optimized neural network and convolutional neural network for automated brain tumor detection

Purpose The purpose of this paper is to develop an efficient brain tumor detection model using the beneficial concept of hybrid classification using magnetic resonance imaging (MRI) images Brain tumors are the most familiar and destructive disease, resulting to a very short life expectancy in their highest grade. The knowledge and the sudden progression in the area of brain imaging technologies have perpetually ready for an essential role in evaluating and concentrating the novel perceptions of brain anatomy and operations. The system of image processing has prevalent usage in the part of medical science for enhancing the early diagnosis and treatment phases. Design/methodology/approach The proposed detection model involves five main phases, namely, image pre-processing, tumor segmentation, feature extraction, third-level discrete wavelet transform (DWT) extraction and detection. Initially, the input MRI image is subjected to pre-processing using different steps called image scaling, entropy-based trilateral filtering and skull stripping. Image scaling is used to resize the image, entropy-based trilateral filtering extends to eradicate the noise from the digital image. Moreover, skull stripping is done by Otsu thresholding. Next to the pre-processing, tumor segmentation is performed by the fuzzy centroid-based region growing algorithm. Once the tumor is segmented from the input MRI image, feature extraction is done, which focuses on the first-order and higher-order statistical measures. In the detection side, a hybrid classifier with the merging of neural network (NN) and convolutional neural network (CNN) is adopted. Here, NN takes the first-order and higher-order statistical measures as input, whereas CNN takes the third level DWT image as input. As an improvement, the number of hidden neurons of both NN and CNN is optimized by a novel meta-heuristic algorithm called Crossover Operated Rooster-based Chicken Swarm Optimization (COR-CSO). The AND operation of outcomes obtained from both optimized NN and CNN categorizes the input image into two classes such as normal and abnormal. Finally, a valuable performance evaluation will prove that the performance of the proposed model is quite good over the entire existing model. Findings From the experimental results, the accuracy of the suggested COR-CSO-NN + CNN was seemed to be 18% superior to support vector machine, 11.3% superior to NN, 22.9% superior to deep belief network, 15.6% superior to CNN and 13.4% superior to NN + CNN, 11.3% superior to particle swarm optimization-NN + CNN, 9.2% superior to grey wolf optimization-NN + CNN, 5.3% superior to whale optimization algorithm-NN + CNN and 3.5% superior to CSO-NN + CNN. Finally, it was concluded that the suggested model is superior in detecting brain tumors effectively using MRI images. Originality/value This paper adopts the latest optimization algorithm called COR-CSO to detect brain tumors using NN and CNN. This is the first study that uses COR-CSO-based optimization for accurate brain tumor detection.

Smart Wristband-Based Stress Detection Framework for Older Adults With Cortisol as Stress Biomarker

In this work, our objective is to design, develop, and evaluate the effectiveness of a stress detection model for older adults using a system of wrist-worn sensors. Our system uses four signals, EDA, BVP, IBI, and ST from EDA, PPG, and ST sensors, embedded in a smart wristband, to classify between stressed and not-stressed state. The stress reference is obtained from salivary cortisol measurement, which is a well established clinical biomarker for measuring physiological stress. This work is the result of year-long data collection and analysis of 40 older adults (28 females and 12 males) and age 73.625 ± 5.39. EDA, BVP, IBI, and ST signals were collected during TSST (Trier Social Stress Test), which is a well known experimental protocol to reliably induce stress in humans in a social setting. 47 features were extracted from EDA, BVP, IBI, and ST signals, out of which 27 features were selected using a supervised feature selection method. Results and analysis show that combining the features from all the four signal streams increases the model's ability to accurately distinguish between the stressed and not-stressed states. The proposed model achieved a macro-average F1-score of 0.92 and an accuracy of 94% in distinguishing between the two states when features from all the four signals were used. Further, we prototype the proposed stress detection model in a consumer end device with voice capabilities, so that users can receive feedback on their vitals and stress levels by querying on voice-enabled consumer devices such as smartphones and smart speakers.

A pliant parametric detection model for line transect data sampling

Line transect survey methodology is a commonly used method for estimating the population abundance. Despite recent advances in this regard, parametric models are still widely used among biometricians, mainly because of their simplicity. In this paper, a new two-parameter detection model satisfying the shoulder conditions is proposed for modeling line transect data. We discuss its properties of interest, including the shapes of the model and the corresponding probability density function, moments, and the related sub-detection model. Maximum likelihood estimation of the parameters is considered. Subsequently, an application is carried out to the proposed model based on a practical data set of perpendicular distances. It is compared with some classical and recent models based on the evaluation of some goodness-of-fit statistics. As results, the variance-covariance matrix, confidence intervals of the parameters and estimated population abundance of the data set are obtained under the proposed detection model.

Optimally Configured Deep Convolutional Neural Network for Attack Detection in Internet of Things: Impact of Algorithm of the Innovative Gunner

Nowadays, the internet of things (IoT) has gained significant research attention. It is becoming critically imperative to protect IoT devices against cyberattacks with the phenomenal intensification. The malicious users or attackers might take control of the devices and serious things will be at stake apart from privacy violation. Therefore, it is important to identify and prevent novel attacks in the IoT context. This paper proposes a novel attack detection system by interlinking the development and operations framework. This proposed detection model includes two stages such as proposed feature extraction and classification. The preliminary phase is feature extraction, the data from every application are processed by integrating the statistical and higher-order statistical features together with the extant features. Based on these extracted features the classification process is evolved for this, an optimized deep convolutional neural network (DCNN) model is utilized. Besides, the count of filters and filter size in the convolution layer, as well as the activation function, are optimized using a new modified algorithm of the innovative gunner (MAIG), which is the enhanced version of the AIG algorithm. Finally, the proposed work is compared and proved over other traditional works concerning positive and negative measures as well. The experimental outcomes show that the proposed MAIG algorithm for application 1 under the GAF-GYT attack achieves higher accuracy of 64.52, 2.38 and 3.76% when compared over the methods like DCNN, AIG and FAE-GWO-DBN, respectively.

Botnet detection using negative selection algorithm, convolution neural network and classification methods

Botnet is a network and internet risk. It is necessary to detect botnet by analyzing and monitoring in order to quickly prevent them. Most approaches are proposed to detect bots using processing and preprocessing on a large number of incoming information from network packets, structures, etc. The recent growth of Internet and network environments has caused a significant growth in botnet attack. Accordingly, the traditional approaches are not good for botnet detection. This paper presents a new approach for the detection of botnet within networks. The proposed detection model is used to compare four attacks, the IRC, HTTP, DNS and P2P, which are used by botnet. Additionally, this model evaluates the accuracy of botnet detection. We use network nerves and correlation and also NSA (negative selection algorithm) which is based on the artificial immune system to identify botnet and compare our results with random forest, K-neighbors, SVM, Gaussian NB, CNN, LSTM algorithms. Our method (CNN-LSTM) presents shorter training time and higher accuracy. In this experiment, we use ISOT and ISCX botnet dataset which are labeled as traffic data. In addition, we investigate various types of botnet attacks and the final evaluation is presented.

Identification and Localization of Myocardial Infarction Based on Analysis of ECG Signal in Cross Spectral Domain Using Boosted SVM Classifier

Automated and accurate detection of myocardial infarction (MI) is a momentous task because of its association with damage of heart muscles and sudden cardiac arrest. This work highlights the investigation of discriminative feature representations to distinguish MI episodes from normal sinus rhythms by combining the strength of boosted classification approach. Proposed method analyzes the spectral coherency of 12 lead electrocardiogram (ECG) signal to explore the discrimination of normal and MI data in cross spectral domain. Coherence spectrum reveals the distinguishing characteristics of the pathologically varying ECG signals which in turn identify and localize the MI episode. In this view, two new feature metrics namely spectral coherence indices (SCIs) and phase coherence indices (PCIs) are introduced to capture the significant coherence variations. The effectiveness of the extracted features is determined by the Fisher score (F-score) estimation. Finally, a boosting framework using the ensemble of support vector machine (SVM) classifiers is presented to improve the performances of the proposed detection model. The boosted SVM augments the boosting technique by introducing a new cluster-based training sample set selection algorithm. The proposed detection model is empirically evaluated using the open access Physikalisch-Technische Bundesanstalt (PTB) diagnostic ECG database. The accuracy of 98.96% and 98.85% is obtained for detecting and identifying the different types of MI pathology, respectively. Comparative study with the existing work exhibits encouraging performances with reduced feature dimension.

DeepfakeNet, an Efficient Deepfake Detection Method

Different CNNs models do not perform well in deepfake detection in cross datasets. This paper proposes a deepfake detection model called DeepfakeNet, which consists of 20 network layers. It refers to the stacking idea of ResNet and the split-transform-merge idea of Inception to design the network block structure, That is, the block structure of ResNeXt. The study uses some data of FaceForensics++, Kaggle and TIMIT datasets, and data enhancement technology is used to expand the datasets for training and testing models. The experimental results show that, compared with the current mainstream models including VGG19, ResNet101, ResNeXt50, XceptionNet and GoogleNet, in the same dataset and preset parameters, the proposed detection model not only has higher accuracy and lower error rate in cross dataset detection, but also has a significant improvement in performance.

Mining Bytecode Features of Smart Contracts to Detect Ponzi Scheme on Blockchain

The emergence of smart contracts has increased the attention of industry and academia to blockchain technology, which is tamper-proofing, decentralized, autonomous, and enables decentralized applications to operate in untrustworthy environments. However, these features of this technology are also easily exploited by unscrupulous individuals, a typical example of which is the Ponzi scheme in Ethereum. The negative effect of unscrupulous individuals writing Ponzi scheme-type smart contracts in Ethereum and then using these contracts to scam large amounts of money has been significant. To solve this problem, we propose a detection model for detecting Ponzi schemes in smart contracts using bytecode. In this model, our innovation is shown in two aspects: We first propose to use two bytes as one characteristic, which can quickly transform the bytecode into a high-dimensional matrix, and this matrix contains all the implied characteristics in the bytecode. Then, We innovatively transformed the Ponzi schemes detection into an anomaly detection problem. Finally, an anomaly detection algorithm is used to identify Ponzi schemes in smart contracts. Experimental results show that the proposed detection model can greatly improve the accuracy of the detection of the Ponzi scheme contracts. Moreover, the F1-score of this model can reach 0.88, which is far better than those of other traditional detection models.

Impact of Code Deobfuscation and Feature Interaction in Android Malware Detection

With more than three million applications already in the Android marketplace, various malware detection systems based on machine learning have been proposed to prevent attacks from cybercriminals; most of these systems use static analyses to extract application features. However, many features generated by static analyses can be easily thwarted by obfuscation techniques. Therefore, several researchers have addressed this obfuscation problem with obfuscation-invariant features. However, to the best of our knowledge, no researcher has utilized deobfuscation techniques. To this end, we adopt a code deobfuscation technique with an Android malware detection system and investigate its effects. Experimental results indicate that code deobfuscation can successfully retrieve useful information concealed by obfuscation. Further, we propose interaction terms based on identified feature interactions. The proposed interaction terms aim to eliminate the interference caused by the size of the application and other features because many feature values are correlated to the size of the application. In addition, the experimental results indicate that these interaction terms have a high ranking in terms of feature importance values. Our proposed Android malware detection model achieves 99.55% accuracy and a 94.61% F1-score with the well-known Drebin dataset, which is better than the performance of previous works.

Botnet Detection Approach Using Graph-Based Machine Learning

Detecting botnet threats has been an ongoing research endeavor. Machine Learning (ML) techniques have been widely used for botnet detection with flow-based features. The prime challenges with flow-based features are that they have high computational overhead and do not fully capture network communication patterns. Recently, graph-based ML has witnessed a dramatic increase in attention. In communication networks, graph data offers insights information about communication patterns between hosts. In this paper, we propose a graph-based ML model for botnet detection that first considers the significance of graph features before developing a generalized model for detecting botnets based on the selected important features. We explore different feature sets selected using five filter-based feature evaluation measures derived from various theories such as consistency, correlation, and information. Two heterogeneous botnet datasets, CTU-13 and IoT-23, were used to evaluate the effectiveness of the proposed graph-based botnet detection with several supervised ML algorithms. Experiment results show that using features reduces training time and model complexity and provides high bots detection rate. Our proposed detection model detects different types of botnet families and exhibits robustness to zero-day attacks. Compared to state-of-the-art techniques flow-, and graph-based, our approach achieves higher precision and shows competitive accuracy.

Research on Sheep Recognition Algorithm Based on Deep Learning in Animal Husbandry

At this stage, the domestic animal husbandry industry has a very important support role in the domestic market economy, and the sheep industry has received extensive attention as one of the important industries. Limb movements and activities can directly reflect the adaptability of sheep to the breeding environment and conditions, and provide a richer scientific and technical experience for sheep breeding. The sheep target detection is an important prerequisite for grasping the movement behavior of sheep. Therefore, how to efficiently and accurately identify and detect sheep has become the key to the development of sheep industry at this stage. In this paper, the Faster-RCNN neural network model based on the Soft-NMS algorithm is studied, which realizes the real-time detection and positioning of sheep under complex breeding conditions, and improves the accuracy of recognition while ensuring the detection speed. Experiments show that the proposed detection model can detect sheep with 95.32% accuracy and mark the location of the target in real time, which provides an effective data foundation for sheep behavior research and helps promote the development of high-tech animal husbandry effect.