GLE-Net: A Global and Local Ensemble Network for Aerial Object Detection

Recent advances in camera-equipped drone applications increased the demand for visual object detection algorithms with deep learning for aerial images. There are several limitations in accuracy for a single deep learning model. Inspired by ensemble learning can significantly improve the generalization ability of the model in the machine learning field, we introduce a novel integration strategy to combine the inference results of two different methods without non-maximum suppression. In this paper, a global and local ensemble network (GLE-Net) was proposed to increase the quality of predictions by considering the global weights for different models and adjusting the local weights for bounding boxes. Specifically, the global module assigns different weights to models. In the local module, we group the bounding boxes that corresponding to the same object as a cluster. Each cluster generates a final predict box and assigns the highest score in the cluster as the score of the final predict box. Experiments on benchmarks VisDrone2019 show promising performance of GLE-Net compared with the baseline network.

An Influence Network-Based Consensus Model for Large-Scale Group Decision Making with Linguistic Information

The vast majority of the existing social network-based group decision-making models require extra information such as trust/distrust, influence and so on. However, in practical decision-making process, it is difficult to get additional information apart from opinions of decision makers. For large-scale group decision making (LSGDM) problem in which decision makers articulate their preferences in the form of comparative linguistic expressions, this paper proposes a consensus model based on an influence network which is inferred directly from preference information. First, a modified agglomerative hierarchical clustering algorithm is developed to detect subgroups in LSGDM problem with flexible linguistic information. Meanwhile, a measure method of group consensus level is proposed and the optimal clustering level can be determined. Second, according to the preference information of group members, influence network is constructed by determining intra-cluster and inter-cluster influence relationships. Third, a two-stage feedback mechanism guided by influence network is established for the consensus reaching process, which adopts cluster adjustment strategy and individual adjustment strategy depending on the different levels of group consensus. The proposed mechanism can not only effectively improve the efficiency of consensus reaching of LSGDM, but also take individual preference adjustment into account. Finally, the feasibility and effectiveness of the proposed method are verified by the case of intelligent environmental protection project location decision.

A Novel Design of Morlet Wavelet to Solve the Dynamics of Nervous Stomach Nonlinear Model

The present study introduces a novel design of Morlet wavelet neural network (MWNN) models to solve a class of a nonlinear nervous stomach system represented with governing ODEs systems via three categories, tension, food and medicine, i.e., TFM model. The comprehensive detail of each category is designated together with the sleep factor, food rate, tension rate, medicine factor and death rate are also provided. The computational structure of MWNNs along with the global search ability of genetic algorithm (GA) and local search competence of active-set algorithms (ASAs), i.e., MWNN-GA-ASAs is applied to solve the TFM model. The optimization of an error function, for nonlinear TFM model and its related boundary conditions, is performed using the hybrid heuristics of GA-ASAs. The performance of the obtained outcomes through MWNN-GA-ASAs for solving the nonlinear TFM model is compared with the results of state of the article numerical computing paradigm via Adams methods to validate the precision of the MWNN-GA-ASAs. Moreover, statistical assessments studies for 50 independent trials with 10 neuron-based networks further authenticate the efficacy, reliability and consistent convergence of the proposed MWNN-GA-ASAs.

Self-organizing Maps and Bayesian Regularized Neural Network for Analyzing Gasoline and Diesel Price Drifts

Any nation’s growth depends on the trend of the price of fuel. The fuel price drifts have both direct and indirect impacts on a nation’s economy. Nation’s growth will be hampered due to the higher level of inflation prevailing in the oil industry. This paper proposed a method of analyzing Gasoline and Diesel Price Drifts based on Self-organizing Maps and Bayesian regularized neural networks. The US gasoline and diesel price timeline dataset is used to validate the proposed approach. In the dataset, all grades, regular, medium, and premium with conventional, reformulated, all formulation of gasoline combinations, and diesel pricing per gallon weekly from 1995 to January 2021, are considered. For the data visualization purpose, we have used self-organizing maps and analyzed them with a neural network algorithm. The nonlinear autoregressive neural network is adopted because of the time series dataset. Three training algorithms are adopted to train the neural networks: Levenberg-Marquard, scaled conjugate gradient, and Bayesian regularization. The results are hopeful and reveal the robustness of the proposed model. In the proposed approach, we have found Levenberg-Marquard error falls from − 0.1074 to 0.1424, scaled conjugate gradient error falls from − 0.1476 to 0.1618, and similarly, Bayesian regularization error falls in − 0.09854 to 0.09871, which showed that out of the three approaches considered, the Bayesian regularization gives better results.

Artificial Electric Field Algorithm with Greedy State Transition Strategy for Spherical Multiple Traveling Salesmen Problem

The multiple traveling salesman problem (MTSP) is an extension of the traveling salesman problem (TSP). It is found that the MTSP problem on a three-dimensional sphere has more research value. In a spherical space, each city is located on the surface of the Earth. To solve this problem, an integer-serialized coding and decoding scheme was adopted, and artificial electric field algorithm (AEFA) was mixed with greedy strategy and state transition strategy, and an artificial electric field algorithm based on greedy state transition strategy (GSTAEFA) was proposed. Greedy state transition strategy provides state transition interference for AEFA, increases the diversity of population, and effectively improves the accuracy of the algorithm. Finally, we test the performance of GSTAEFA by optimizing examples with different numbers of cities. Experimental results show that GSTAEFA has better performance in solving SMTSP problems than other swarm intelligence algorithms.

Daily Activity Recognition and Tremor Quantification from Accelerometer Data for Patients with Essential Tremor Using Stacked Denoising Autoencoders

Human activity recognition (HAR) has received more and more attention, which is able to play an important role in many fields, such as healthcare and intelligent home. Thus, we have discussed an application of activity recognition in the healthcare field in this paper. Essential tremor (ET) is a common neurological disorder that can make people with this disease rise involuntary tremor. Nowadays, the disease is easy to be misdiagnosed as other diseases. We have combined the essential tremor and activity recognition to recognize ET patients’ activities and evaluate the degree of ET for providing an auxiliary analysis toward disease diagnosis by utilizing stacked denoising autoencoder (SDAE) model. Meanwhile, it is difficult for model to learn enough useful features due to the small behavior dataset from ET patients. Thus, resampling techniques are proposed to alleviate small sample size and imbalanced samples problems. In our experiment, 20 patients with ET and 5 healthy people have been chosen to collect their acceleration data for activity recognition. The experimental results show the significant result on ET patients activity recognition and the SDAE model has achieved an overall accuracy of 93.33%. What’s more, this model is also used to evaluate the degree of ET and has achieved the accuracy of 95.74%. According to a set of experiments, the model we used is able to acquire significant performance on ET patients activity recognition and degree of tremor assessment.

Application of Improved Differential Evolution Algorithm in Solving Equations

The problem of finding roots of equations has always been an important research problem in the fields of scientific and engineering calculations. For the standard differential evolution algorithm cannot balance the convergence speed and the accuracy of the solution, an improved differential evolution algorithm is proposed. First, the one-half rule is introduced in the mutation process, that is, half of the individuals perform differential evolutionary mutation, and the other half perform evolutionary strategy reorganization, which increases the diversity of the population and avoids premature convergence of the algorithm; Second, set up an adaptive mutation operator and a crossover operator to prevent the algorithm from falling into the local optimum and improve the accuracy of the solution. Finally, classical high-order algebraic equations and nonlinear equations are selected for testing, and compared with other algorithms. The results show that the improved algorithm has higher solution accuracy and robustness, and has a faster convergence speed. It has outstanding effects in finding roots of equations, and provides an effective method for engineering and scientific calculations.

Network Intrusion Detection: An Analytical Assessment Using Deep Learning and State-of-the-Art Machine Learning Models

The internet connected devices are prone to cyber threats. Most of the companies are developing devices with built-in cyber threat protection mechanism or recommending prevention measure. But cyber threat is becoming harder to trace due to the availability of various tools and techniques to bypass the normal prevention measures. A data mining-based intrusion detection system can play a key role to handle such cyberattacks. This paper proposes a threefold approach to analyzing intrusion detection system. In the first phase, experiments have been conducted by applying SVM, Decision Tree, and KNN. In the second phase, Random Forest, and XGBoost are applied as lately they have been showing significant improved performance in supervised learning. Finally, deep learning techniques, namely, Feed Forward, LSTM, and Gated Recurrent Unit neural network are applied to conduct the experiment. Kyoto Honeypot Dataset is used for experimental purpose. The results show a significant improvement in IDS outperforming the state of the arts on this dataset. Such improvement strengthens the applicability proposed model in IDS.

Research on Multi-label Text Classification Method Based on tALBERT-CNN

Single-label classification technology has difficulty meeting the needs of text classification, and multi-label text classification has become an important research issue in natural language processing (NLP). Extracting semantic features from different levels and granularities of text is a basic and key task in multi-label text classification research. A topic model is an effective method for the automatic organization and induction of text information. It can reveal the latent semantics of documents and analyze the topics contained in massive information. Therefore, this paper proposes a multi-label text classification method based on tALBERT-CNN: an LDA topic model and ALBERT model are used to obtain the topic vector and semantic context vector of each word (document), a certain fusion mechanism is adopted to obtain in-depth topic and semantic representations of the document, and the multi-label features of the text are extracted through the TextCNN model to train a multi-label classifier. The experimental results obtained on standard datasets show that the proposed method can extract multi-label features from documents, and its performance is better than that of the existing state-of-the-art multi-label text classification algorithms.

A Manufacturing SCOS Model (MSCOS) Based on the Similarity of Parameter Sequences Between Tasks and Service Composition

Service composition and optimal selection (SCOS) is a core issue in cloud manufacturing (CMfg) when integrating distributed manufacturing services for complex manufacturing tasks. Generally, a set of recommended task parameter sequences (Tps) will be given when publishing manufacturing tasks. The similarity between the service composition parameter sequence (SCps) and Tps also reflects the rationality of the service composition. However, various evaluation models based on QoS have been proposed, ignoring the rationality between the Tps and SCps. Considering the similarity of the Tps and SCps in an evaluation model, we propose a manufacturing SCOS framework called MSCOS. The framework includes two parts: an evaluation model and an algorithm for both optimization and selection. In the evaluation model, based on the numerical proximity and geometric similarity between the Tps and SCps, improving the technique for order preference by similarity to an ideal solution (TOPSIS) with the grey correlation degree (GC), we propose the GC&TOPSIS (GTOPSIS). In the optimization and selection algorithm, an improved flower pollination algorithm (IFPA) is proposed to achieve optimization and selection based on polyline characteristics between the fitness values in the population. Experiments show that the MSCOS evaluation effect and optimal selection offer better performance than commonly used algorithms.