Electric Equipment Image Recognition Based on Deep Learning and Random Forest

Deep learning is a complex machine learning algorithm that can learn the deep inner rules of a sample. In recent years, deep learning has achieved good results in the field of image processing. The essence of deep learning is to build a multi-layered machine learning architecture that trains on large-scale data. With the development of computer technology, deep learning has achieved good results in image recognition and natural language processing. Deep learning is also becoming more and more widely used in industry. This paper introduces the applications of deep learning in the industrial field, and finally discusses development and challenges of deep learning in industrial applications in the future.

Objective To preliminarily analyze the prediction efficiency of multimodal MRI-based radiomics model for preoperative glioma IDH1 gene expression type. Methods The MRI data of 108 surgery-proven glioma patients from May 2015 to January 2019 were retrospectively analyzed, and the MRI data included axial T1WI,T2WI,fluid attenuated inversion recovery (FLAIR),DWI imaging and enhanced T1WI sequence.Forty-seven cases were IDH1 mutant type, and 61 cases were IDH1 wild type. All patients were divided into training and validation groups according to the 7∶3 ratio of the random forest model. Seventy-three cases were in training group, and 35 cases were in validation group. Independent predictors of IDH1 mutation were screened by univariate analysis combined with multivariate logistic regression analysis (P<0.05) in order to construct a random forest diagnosis model of general clinical information and conventional MRI morphological features.General clinical information and conventional MRI morphological features included gender, age, umbers of cases of left and right hemispheres, location of tumors, maximum diameter of tumors, peritumoral edema, intratumoral cystic degeneration, enhancement and ADC value of tumors. The ROI was manually outlined by MaZda software in the most obvious level of 5 sequences of tumor mass and the radiomics features were extracted, including the gray-level co-occurrence matrix(GLCM), the run-length matrix(RUN), the absolute gradient(GRA),the auto-regressive model(ARM) and wavelets transform(WAV). The least absolute shrinkage and selection operator (LASSO)regression were used to select image radiomics features with a method of 10 fold cross -validation and to reduce the dimensions. The screened image radiomics labels were combined with the conventional morphological feature independent predictors to construct a multimodal MRI-based random forest model, and the validation data set was used to evaluate the accuracy and diagnostic efficiency of each model. The sensitivity and specificity of conventional MRI morphological feature model and multimodal MRI-based radiomics prediction model were evaluated dynamically by drawing ROC curves, and the prediction efficiency of the two models was quantified by using AUC statistical indicators. The model classification error rate under different outcomes and the classification error rate of out of bag(OOB)were used to evaluate the stability of the multimodal MRI-based random forest model. The contribution rate of each variable to the model was reflected by the characteristic variables importance assessment map. Results Univariate regression analysis of the conventional MRI morphological characteristics showed that peritumoral edema, cystic degeneration and enhancement were the three independent predictors of IDH1 gene expression (P<0.01). LASSO algorithm and 10-fold cross-validation identified six robust radiomic features including high frequency coefficients of wavelet transform (WavEnHH_s-4) of T2WI, S(4,4) inverse difference of gray uniformity measurement (InvDfMom),S(5,0) Entropy (entropy),WavEnHH_s-4 of T1WI enhancement, S(1,1) InvDfMom,S(1, -1) Entropy Difference (DifEntrp)of Flair.The error rate of classification for different outcomes and classification error rate of random forest OOB data of multimodal MRI radiomics diagnosis model finally stabilized at 10%. The results of Characteristic Variable Importance Assessment Map: Mean Decrease Accuracy and Mean Decrease Gini index were consistent, which showed that besides three conventional MRI morphological predictors peritumoral edema, enhancement and cystic degeneration, the radiomics labels also played a key role in the model. The results of ROC curve showed that the accuracy, specificity, sensitivity and AUC of conventional MRI morphological feature model were 82.7%, 68.4%, 90.9% and 0.835, respectively, and those of multimodal MRI-based radiomics model were 88.5%, 89.5%, 87.8% and 0.956 respectively. Conclusion Multimodal MRI-based radiomics random forest model can improve the predictive efficiency of preoperative glioma IDH1 gene expression type more quantitatively. Key words: Glioma; Texture; Radiomics; Magnetic resonance imaging

Active learning is one form of supervised machine learning. In supervised learning, a set of labeled samples is passed to a learning algorithm for training a classifier. However, labeling large amounts of training samples can be costly and error-prone. Active learning deals with the development of algorithms that interactively select a subset of the available unlabeled samples for labeling, and aims at minimizing the labeling effort while maintaining classification performance. The key challenge for the development of so-called active learning strategies is the balance between exploitation and exploration: On the one hand, the estimated decision boundary needs to be refined in feature space regions where it has already been established, while, on the other hand, the feature space needs to be scanned carefully for unexpected class distributions. In this thesis, two approaches to active learning are presented that consider these two aspects in a novel way. In order to lay the foundations for the first one, it is proposed to express the uncertainty in class prediction of a classifier at a test point in terms of a second-order distribution. The mean of this distribution corresponds to the common estimate of the posterior class probabilities and thus is related to the distance of the test point to the decision boundary, whereas the spread of the distribution indicates the degree of exploration in the corresponding region of feature space. This allows for the evaluation of the utility of labeling a yet unlabeled point with respect to classifier improvement in a principled way and leads to a completely novel approach to active learning. The proposed strategy is then implemented and evaluated based on kernel density classification. The generic active learning strategy can be combined with any other classifier, but it performs best if the derived second-order distributions are sufficiently good approximations to the sampling distribution. Although second-order distributions for random forests are derived in this thesis, they do not approximate sufficiently well the sampling distribution and mainly allow only for the relative comparison of prediction uncertainty between test points. In order to combine the state of the art classification performance of random forests with the principal ideas of the first active learning approach, a related second approach for random forests is derived. It is, in addition, tailored to the demands in industrial optical inspection: bag-wise labeling with weak labels and strongly imbalanced classes. Moreover, an outlier detection scheme based on random forests is derived that is used by the proposed active learning algorithm. Finally, a new computational scheme for Gaussian process classification is presented. It is compared to two standard methods in geostatistics, both with respect to theoretical consistency and practical performance. The method evolved as a by-product when considering using Gaussian process models for active learning.

People today, in addition to their concerns about getting old and having to go through watching themselves grow weak and wrinkly, are facing an increasing fear of dementia. There are around 47 million people affected by dementia worldwide and the cost associated with providing them health and social care support is estimated to reach 2 trillion by 2030 which is almost equivalent to the 18th largest economy in the world. The most common form of dementia with the highest costs in health and social care is Alzheimer’s disease, which gradually kills neurons and causes patients to lose loving memories, the ability to recognise family members, childhood memories, and even the ability to follow simple instructions. Alzheimer’s disease is irreversible, unstoppable and has no known cure. Besides being a calamity to affected patients, it is a great financial burden on health providers. Health care providers also face a challenge in diagnosing the disease as current methods used to diagnose Alzheimer’s disease rely on manual evaluations of a patient’s medical history and mental examinations such as the Mini-Mental State Examination. These diagnostic methods often give a false diagnosis and were designed to identify Alzheimer’s after stage two when the part of all symptoms are evident. The problem is that clinicians are unable to stop or control the progress of Alzheimer’s disease, because of a lack of knowledge on the patterns that triggered the development of the disease. In this thesis, we explored and investigated Alzheimer’s disease from a computational perspective to uncover different risk factors and present a strategic framework called Early Prediction of Alzheimer’s Disease Framework (EPADf) that would give a future prediction of early-onset Alzheimer’s disease. Following extensive background research that resulted in the formalisation of the framework concept, prediction approaches, and the concept of ranking the risk factors based on clinical instinct, knowledge and experience using mathematical reasoning, we carried out experiments to get further insight and investigate the disease further using machine learning models. In this study, we used machine learning models and conducted two classification experiments for early prediction of Alzheimer’s disease, and one ranking experiment to rank its risk factors by importance. Besides these experiments, we also presented two logical approaches to search for patterns in an Alzheimer’s dataset, and a ranking algorithm to rank Alzheimer’s disease risk factors based on clinical evaluation. For the classification experiments we used five different Machine Learning models; Random Forest (RF), Random Oracle Model (ROM), a hybrid model combined of Levenberg-Marquardt neural network and Random Forest, combined using Fischer discriminate analysis (H2), Linear Neural Networks (LNN), and Multi-layer Perceptron Model (MLP). These models were deployed on a de-identified multivariable patient’s data, provided by the ADNI (Alzheimer’s disease Neuroimaging Initiative), to illustrate the effective use of data analysis to investigate Alzheimer’s disease biological and behavioural risk factors. We found that the continues enhancement of patient’s data and the use of combined machine learning models can provide an early cost-effective prediction of Alzheimer’s disease, and help in extracting insightful information on the risk factors of the disease. Based on this work and findings we have developed the strategic framework (EPADf) which is discussed in more depth in this thesis.

Caveolae are plasma membrane invaginations whose formation requires caveolin-1 (Cav1), the adaptor protein polymerase I, and the transcript release factor (PTRF or CAVIN1). Caveolae have an important role in cell functioning, signaling, and disease. In the absence of CAVIN1/PTRF, Cav1 forms non-caveolar membrane domains called scaffolds. In this work, we train machine learning models to automatically distinguish between caveolae and scaffolds from single molecule localization microscopy (SMLM) data. We apply machine and deep learning algorithms to discriminate biological structures from SMLM data. Our work is the first that is leveraging machine and deep learning approaches to automatically identifying biological structures from SMLM data. In particular, we develop and compare three binary classification methods to identify whether or not a given 3D cluster of Cav1 proteins is a caveolae. The first uses a random forest classifier applied to 28 hand-picked features, the second uses a convolutional neural net (CNN) applied to a projection of the point clouds onto three planes, and the third uses a PointNet model, a recent development that can directly take point clouds as its input. We validate our methods on a dataset of super-resolution microscopy images of PC3 prostate cancer cells labeled for Cav1. Specifically, we have images from two cell populations: 10 PC3 and 10 CAVIN1/PTRF-transfected PC3 cells (PC3-PTRF cells) that form caveolae. We obtained a balanced set of 1714 different cellular structures. Our results show that both the random forest on hand selected features and the deep learning approach achieve high accuracy in distinguishing the intrinsic features of the caveolae and non-caveolae biological structures. More specifically, both random forest and deep CNN classifiers achieve classification accuracy reaching 94% on our test set, while the PointNet model only reached 83% accuracy. We also discuss the pros and cons of the different approaches. Additional information about individual files is in the accompanying CSV file (item_metadata.csv). This dataset was originally deposited in the Simon Fraser University institutional repository.

Necrotising enterocolitis (NEC) is an inflammatory intestinal disorder affecting premature infants. Despite the worldwide improvement of health care practices and facilities raising the survival rates of neonatal premature infants, there has not been any improvement in treatment options or mortality rates for NEC. There has been an extensive volume of research into NEC, though to date there has not been any evidence to directly associate a causal agent to this devastating disease, nor have there been any conclusive observations of NEC prior to birth. The only key prognostic signal for NEC is that onset and severity of the disease are significantly associated with the prematurity of the neonatal infant. During the process of birth, the infant transitions from the near sterile conditions of the womb to the outside environment teeming with bacteria. Only then does the infant develop a symbiotic relationship as a host to beneficial bacteria. Upon this transition the community of gut microbes develops and aids in a range of host functions, namely digestion and absorption of nutrients, as well as the immune response. It is also at this time that NEC can begin to develop in the lower gastrointestinal tract. This coinciding of factors implicates the colonisation of the gut with bacteria in the development of NEC, however studies to date have failed to provide consistent reports of a causative pathogen or a characteristic gut microbiome structure associated with NEC. The purpose of this study was to characterise the structure of the bacterial community in the gastrointestinal tract of infants with and without NEC in order to identify a community distinctive to infants with NEC. This was addressed using non-invasive faecal sampling of premature infants in a large, prospectively enrolled cohort from across England, sampled over a twenty-day window spanning ten days prior to and ten days following the onset of NEC. Using established V4 16S rRNA protocols, bacterial taxa within environmental samples could be characterised without relying on classic culture dependent methods. With this amplification technique it was possible to utilise small amounts of bacterial DNA isolated from infant faecal samples while at the same time mitigating the selection bias associated with culture-based techniques. Short read sequencing (illumina MiSeq) was performed on a total of 656 faecal samples from 132 infants spanning eight neonatal intensive care units across England. All infants had gestational durations of less than 35 weeks. 44 infants had NEC (225 samples) and 88 were assigned as controls (431 samples) according to key risk factors defined by medical practitioners. Taxonomic abundances assigned with the QIIME informatics pipeline were normalised using non-reductive negative binomial normalisation. Local contributions to beta-diversity (LCBD) scores were used to quantify taxonomic changes in the community structure through subset regression. Non-metric multidimensional scaling (NMDS) was used to establish risk factors that best described NEC and control samples. The Random Forest machine learning algorithm was used to establish taxa that best discriminated between NEC and control infants, as well as to identify any conserved pathogens. Subset regression identified feeding regime, mode of delivery and age at sampling as significant discriminating factors for the NEC status of infants based on sample LCBD values. However, NMDS plots of sample LCBD values showed no clear clustering of samples according to NEC status. Canonical correlation analysis (CCA) indicated that this variability was due to inter-individual differences. Of the risk factors that could be accounted for, feeding regime was the most effective in differentiating community structures of NEC and control infant samples. There was also evidence that initial communities were influenced by delivery method. Three subgroups of infants based on these influential risk factors and with sufficient sampling depth were established and analysed separately in addition to collective, population-scale analysis. Random Forest analysis demonstrated that reduced abundance of the genus Bifidobacterium was significantly associated with NEC across all sub-groups of infants. Additionally, this method of analysis indicated no clear pathogenic taxa that consistently spanned the population. For infants with NEC that were delivered by caesarean section and fed both formula and breast milk, there was increased abundance of the genus Dialister when sampled shortly after birth. Infants that did not develop NEC, who were delivered vaginally and fed both formula and breast milk were seen to have a significantly greater abundance of the genus Veillonella over time, relative to NEC subjects. There were no additional taxonomic differences that could be ascertained in the sub-group of infants who were vaginally delivered and fed breast milk exclusively. Overall, the unique nature of the microbiome and the high degree of inter-individual variation within the community made direct comparisons between NEC and non-NEC subjects difficult. However, by accounting for factors that were significantly associated with NEC status it was possible to observe consistent association of increased bifidobacterial abundance in infants that did not develop NEC. This highlights the importance of large scale studies and case-control assignment when analysing complex community structures such as that of the human gastrointestinal tract, as well as the powerful, deep analytical analysis provided by machine learning algorithms. Further work should look to establish the impact and role of Bifidobacteria in the human gut community to inform early interventions in healthcare settings for infants at risk of NEC, focussing specifically on encouraging the development of a community structure reflective of that observed in premature infants that do not develop NEC.

Background: From a mathematical point of view, the problems of medical diagnostics are the tasks of data classification. It is important to understand how significant distortions can contribute to the result of classification errors in the collection of primary diagnostic information, in particular, the results of biochemical tests. Aims: Determination of the dependence of the prediction result on the variability of the primary diagnostic information on the example of the model classifier. Materials and methods : The case-control study enrolled patients who were divided into 2 groups: the main (diagnosed with lung cancer, n=200) and the control group (conditionally healthy, n=500). Questioning and biochemical saliva study was performed in all participants. Patients of the main group and the comparison group were hospitalized for surgical treatment, after which carried out the histological verification of the diagnosis. The biochemical composition of saliva is determined spectrophotometrically. Based on the data obtained, a model classifier for the diagnosis of lung cancer (a random forest) has been constructed. In each parameter underlying the classifier, deviations were made in the specified range (±1–5%, ±5–10%, ±10–15%), creating synthetic images. Then, the results of the classification were evaluated by the cross-validation method. Results: The basic diagnostic characteristics of the model classifier are determined (sensitivity ― 72.5%, specificity ― 86.0%). As the deviations of synthetic images from the baseline increase, diagnostic characteristics deteriorate with the general classification. However, the result of a confident classification, on the contrary, gives higher values (sensitivity ― 81.8%, specificity ― 93.1%). In case of a confident classification, similar images that fall into different classes according to the classification results are deleted, whereas in the case of a general classification, they are taken into account. The difference between methods of classification is associated with the presence of images on which the classifier gives the result of belonging to the class in the range of 0.45–0.55. Therefore, it is necessary to introduce a third class into the classifier, the so-called gray zone (0.4–0.6), since the probability of making an erroneous diagnosis in this area is significantly increased. Conclusions: The obtained results allow to conclude that the measurement error in the range (±1–15%) does not significantly affect the quality of the classification.

Machine learning techniques are the cornerstone to handlethe amounts of information available for building comprehensivemodels for decision support in medical practice. However, thedatasets use to have a lot of missing information. In this work weanalyse how the random forests technique could be used for dealingwith missing phenotype values in order to prognosticate diabetestype 2

This thesis considers two main subjects divided in four problems in the broad field of mathematical finance. The first chapter treats option pricing followed by three chapters on the application of the machine learning algorithm of Random Forests to finance, specifically to risk capital aggregation, portfolio optimization and macro stress testing. In all four chapters new methodologies to treat the respective subjects are developed. All proposed models are benchmarked against commonly applied methods in the respective fields and found to outperform their peers.

Image recognition is one of the key technologies for worker’s helmet detection using an unmanned aerial vehicle (UAV). By analyzing the image feature extraction method for workers’ helmet detection based on convolutional neural network (CNN), a double-channel convolutional neural network (DCNN) model is proposed to improve the traditional image processing methods. On the basis of AlexNet model, the image features of the worker can be extracted using two independent CNNs, and the essential image features can be better reflected considering the abstraction degree of the features. Combining a traditional machine learning method and random forest (RF), an intelligent recognition algorithm based on DCNN and RF is proposed for workers’ helmet detection. The experimental results show that deep learning (DL) is closely related to the traditional machine learning methods. Moreover, adding a DL module to the traditional machine learning framework can improve the recognition accuracy.

This paper carries on research surrounding the influences produced by dimensionality reduction on machine learning classification effect. Firstly, paper constructs the analysis architecture of data dimension reduction classification, combines the two different unsupervised dimension reduction methods, locally linear embedding (LLE) and principal component analysis (PCA) with the five machine learning classification methods: Gradient Boosting Decision Tree (GBDT), Random Forest, Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Logistic Regression. And then uses the handwritten digital identification dataset to analyze the classification performance of these five classification methods on different dimension datasets by different dimensionality reduction methods. The analysis shows that using the appropriate dimensionality reduction method for dimensionality reduction classification can effectively improve the classification accuracy; the dimensionality reduction classification effect of non-linear dimensionality reduction method is generally better than the linear dimensionality reduction method; different machine learning classification algorithms have significant differences in the sensitivity of dimensions.

Coronavirus disease (COVID-19) constitutes an ongoing global health problem with significant morbidity and mortality. It usually presents characteristic findings on a chest CT scan, which may lead to early detection of the disease. A timely and accurate diagnosis of COVID-19 is the cornerstone for the prompt management of the patients. The aim of the present study was to evaluate the performance of an automated machine learning algorithm in the diagnosis of Covid-19 pneumonia using chest CT scans. Diagnostic performance was assessed by the area under the receiver operating characteristic curve (AUC), sensitivity, and positive predictive value. The method's average precision was 0.932. We suggest that auto-ML platforms help users with limited ML expertise train image recognition models by only uploading the examined dataset and performing some basic settings. Such methods could deliver significant potential benefits for patients in the future by allowing for earlier disease detection and care.

BackgroundLandslide-affecting factors are uncorrelated or non-linearly correlated, limiting the predictive performance of traditional machine learning methods for landslide susceptibility assessment. Deep learning methods can take advantage of the high-level representation and reconstruction of information from landslide-affecting factors. In this paper, a novel deep learning-based algorithm that combine classifiers of both deep learning and machine learning is proposed for landslide susceptibility assessment. A stacked autoencoder (StAE) and a sparse autoencoder (SpAE) both consist of an input layer for raw data, hidden layer for feature extraction, and output layer for classification and prediction. As a study case, Oda City and Gotsu City in Shimane Prefecture, southwestern Japan, were used for susceptibility assessment and prediction of landslides triggered by extreme rainfall.ResultsThe prediction performance was compared by analyzing real landslide and non-landslide data. The prediction performance of random forest (RF) was evaluated as better than that of a support vector machine (SVM) in traditional machine learning, so RF was combined with both StAE and SpAE. The results show that the prediction ratio of the combined classifiers was 93.2% for StAE combined with RF model and 92.5% for SpAE combined with RF model, which were higher than those of the SVM (87.4%), RF (89.7%), StAE (84.2%), and SpAE (88.2%).ConclusionsThis study provides an example of combined classifiers giving a better predictive ratio than a single classifier. The asymmetric and unsupervised autoencoder combined with RF can exploit optimal non-linear features from landslide-affecting factors successfully, outperforms some conventional machine learning methods, and is promising for landslide susceptibility assessment.

Social media may serve as an important platform for the monitoring of population-level opioid abuse in near real-time. Our objectives for this study were to (i) manually characterize a sample of opioid-mentioning Twitter posts, (ii) compare the rates of abuse/misuse related posts between prescription and illicit opiods, and (iii) to implement and evaluate the performances ofsupervised machine learning algorithms for the characterization of opioid-related chatter, which can potentially automate social media based monitoring in the future.. We annotated a total of 9006 tweets into four categories, trained several machine learning algorithms and compared their performances. Deep convolutional neural networks marginally outperformed support vector machines and random forests, with an accuracy of 70.4%. Lack of context in tweets and data imbalance resulted in misclassification of many tweets to the majority class. The automatic classification experiments produced promising results, although there is room for improvement.

Analytic tools are useful for detecting patterns in education data and providing insights about student performance and learning. This study compared six supervised learning algorithms (linear regression, ridge regression, the lasso, regression trees, random forests regression, gradient boosted regression) and identified features important for predicting student performance. The dataset consisted of N=1044 observations from two secondary schools in Portugal (UCI-MLR, Cortez & Silva, 2008). Performance was assessed by final grades (range: 0-20) in two courses, mathematics and Portugese. The models were fit to training data with 27 independent variables and evaluated on a testing subset. Overall, performance was lower for students in mathematics than Portugese. The models selected a similar set of variables as important for predicting performance: mother's education level, student plans for higher education, and weekly study time were positively related to predicted performance, whereas course subject, school educational support, and romantic relationships were associated with decreased student performance. The models differed in the number, weighting, order and importance given to predictor variables. Linear regression provided a model with 13 predictors. Ridge regression shrank the coefficient estimates toward zero; the lasso performed variables selection for a model with 20 predictors. There was a tradeoff between model complexity and interpretability. The single pruned regression tree provided a simple, interpretable non-linear model with four features. Random forests regression and gradient boosting reduced overfitting, but were more difficult to interpret. Advantages and limitations of the different models are discussed. Applications for educational data mining (EDM) and learning analytics (LA) are considered.