Level Of Classification Accuracy Research Articles

Towards better intelligent implementation of Schizophrenia prediction using federated deep learning framework

Early detection of individuals susceptible to Schizophrenia (SZ) is critical for early intervention, which can reduce the risk of psychosis. This research proposes a deep learning method for classifying EEG data by picking important discriminative EEG features. While existing systems employ an R-CNN methodology, we propose a hybrid CNN–Bi- LSTM automated system that analyses EEG statistical data and performs the prediction. It uses a CNN for an optimised feature selection process to select the most important and informative features, with Bi-LSTM for prediction of susceptibility to develop SZ. The model when run on EEG data of schizophrenic paradigms gives output as ”Schizophrenic” or ”Non-schizophrenic”. This method has a high level of classification accuracy when compared to most existing machine learning models. While it displays a lower accuracy than some complex deep learning systems, it is much more stable and easy to interpret and thus is more practical for clinical settings.

A Transfer Learning Algorithm to Reduce Brain-Computer Interface Calibration Time for Long-Term Users.

Current motor imagery-based brain-computer interface (BCI) systems require a long calibration time at the beginning of each session before they can be used with adequate levels of classification accuracy. In particular, this issue can be a significant burden for long term BCI users. This article proposes a novel transfer learning algorithm, called r-KLwDSA, to reduce the BCI calibration time for long-term users. The proposed r-KLwDSA algorithm aligns the user's EEG data collected in previous sessions to the few EEG trials collected in the current session, using a novel linear alignment method. Thereafter, the aligned EEG trials from the previous sessions and the few EEG trials from the current sessions are fused through a weighting mechanism before they are used for calibrating the BCI model. To validate the proposed algorithm, a large dataset containing the EEG data from 11 stroke patients, each performing 18 BCI sessions, was used. The proposed framework demonstrated a significant improvement in the classification accuracy, of over 4% compared to the session-specific algorithm, when there were as few as two trials per class available from the current session. The proposed algorithm was particularly successful in improving the BCI accuracy of the sessions that had initial session-specific accuracy below 60%, with an average improvement of around 10% in the accuracy, leading to more stroke patients having meaningful BCI rehabilitation.

Feature Set for Classification of Man-Made Underwater Objects in Optical and SAS Data

This paper proposes a feature characterization method that can discriminate between man-made and natural objects on the seabed using a pair of SAS and optical images. Feature matching methods, like scale-invariant feature transform (SIFT), usually utilize gradient information to detect feature points, whereas contour-based features often use geometrical descriptors such as roughness, circularity, and solidity. However, due to the different characteristics of sonar and optical imagery, current descriptors cannot be directly applied over sonar-optical image pairs. The clarity of the water (muddy water, concentration of particles that scatter light) for optical images, and the stability of the platform, and reflections from seabed for SAS images are important parameters that distort the two images and make it hard to match between the two image sources. This paper proposes a feature characterization method that can identify man-made objects in underwater optic-SAS image pairs. Two new contour-based feature descriptors are introduced. The first is the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">entropy angle feature</i> , which aims to reflect the entropy of the angles that originate from points on the contour. The second feature is <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">local curve fitting</i> , which characterizes the radial-polar space of the local curve polynomial fitting. The descriptors are designed to characterize man-made objects whose contour is expected to be smoother than natural objects such as rocks. Experimental results on a real, large dataset comprised of 1,519 optic-SAS image pairs, are shown to verify that the proposed method has a high level of classification accuracy and good discrimination between man-made and natural objects.

LPIN: A Lightweight Progressive Inpainting Network for Improving the Robustness of Remote Sensing Images Scene Classification

At present, the classification accuracy of high-resolution Remote Sensing Image Scene Classification (RSISC) has reached a quite high level on standard datasets. However, when coming to practical application, the intrinsic noise of satellite sensors and the disturbance of atmospheric environment often degrade real Remote Sensing (RS) images. It introduces defects to them, which affects the performance and reduces the robustness of RSISC methods. Moreover, due to the restriction of memory and power consumption, the methods also need a small number of parameters and fast computing speed to be implemented on small portable systems such as unmanned aerial vehicles. In this paper, a Lightweight Progressive Inpainting Network (LPIN) and a novel combined approach of LPIN and the existing RSISC methods are proposed to improve the robustness of RSISC tasks and satisfy the requirement of methods on portable systems. The defects in real RS images are inpainted by LPIN to provide a purified input for classification. With the combined approach, the classification accuracy on RS images with defects can be improved to the original level of those without defects. The LPIN is designed on the consideration of lightweight model. Measures are adopted to ensure a high gradient transmission efficiency while reducing the number of network parameters. Multiple loss functions are used to get reasonable and realistic inpainting results. Extensive tests of image inpainting of LPIN and classification tests with the combined approach on NWPU-RESISC45, UC Merced Land-Use and AID datasets are carried out which indicate that the LPIN achieves a state-of-the-art inpainting quality with less parameters and a faster inpainting speed. Furthermore, the combined approach keeps the comparable classification accuracy level on RS images with defects as that without defects, which will improve the robustness of high-resolution RSISC tasks.

Influence of gray level co-occurrence matrix for texture feature extraction on identification of batik motifs using k-nearest neighbor

Batik is one type of fabric that is unique because it has a special motif, in Indonesia itself batik is unique because it has certain motifs that are made based on the culture from which batik was made. This study aims to examine the effect of the texture feature extraction method on the identification of batik motifs from five major islands in Indonesia. The method used in this study is the Gray Level Co-occurrence Matrix as the texture feature extraction of batik motifs to obtain good batik motif identification accuracy results and to determine the value of the proximity of the training data and image testing of batik motifs, the K-Nearest Neighbor classification method will be used based on texture feature extraction value obtained. In this experiment, 5 experiments will be carried out based on angles 0 degrees, 45 degrees, 90 degrees, 135 degrees, and 180 degrees using the values of k is1, 3, 5, and 7. The confusion matrix will be used to calculate the accuracy level of the K-Nearest Neighbor classification. From the results of experiments carried out using training data as many as 607 images and testing as many as 344 images in five classes used with angles of 0 degrees, 45 degrees, 90 degrees, 135 degrees, 180 degrees, and values of k are 1, 3, 5, and 7, getting the highest accuracy results is at an angle of 135 degrees and 180 degrees with a value of k is 1 of 89.24% and the lowest is at an angle of 90 degrees with a value of k is 3 of 67.44%. This shows that the Gray level co-occurrence matrix method is good for extracting the texture features of batik motifs from five major islands in Indonesia, it is evidenced by the results of the average accuracy of the classification obtained.

Monitoring Perubahan Penggunaan Lahan Pesisir di Kabupaten Gianyar Menggunakan Citra Landsat 1997 dan 2018

The sustainable development process followed by high population growth is an important factor causing land conversion in coastal areas. Analysis of change and planning of land use is one form of coastal resource management to reduce land damage so that it can be used sustainably. This study aims to determine the types and changes in land use over 21-year period (1998-2018) in the coastal area of Gianyar Regency. This study also wants to examine the level of accuracy of Landsat imagery in land use classifications. The Landsat imagery used in this study was Landsat 7 ETM+ in 1998 and Landsat 8 Oli in 2018. The classification method chosen was the supervised classification with the Maximum Likelihood method. The software used is SAGA GIS to process images and QGIS to design map layouts. The results showed that the land use classes in the study area were water, bare land, settlements and buildings, shrubs/bushes, rice fields, moor and mixed gardens. In the period 1998-2018, four classes experienced declined in area, namely water (1.71 ha), bare land (54.99 ha), mixed gardens (28.08 ha), and rice fields (226.62 ha). Meanwhile, land use classes that have increased in area are shrubs/bushes (44.28 ha), moor (104.58 ha), and settlements and buildings (162.54 ha). The overall level of accuracy in land use classification is 85%.

SWANN: Small-World Architecture for Fast Convergence of Neural Networks

On-device intelligence has become increasingly widespread in the modern smart application landscape. A standing challenge for the applicability of on-device intelligence is the excessively high computation cost of training highly accurate Deep Learning (DL) models. These models require a large number of training iterations to reach a high convergence accuracy, hindering their applicability to resource-constrained embedded devices. This paper proposes a novel transformation which changes the topology of the DL architecture to reach an optimal cross-layer connectivity. This, in turn, significantly reduces the number of training iterations required for reaching a target accuracy. Our transformation leverages the important observation that for a set level of accuracy, convergence is fastest when network topology reaches the boundary of a Small-World Network. Small-world graphs are known to possess a specific connectivity structure that enables enhanced signal propagation among nodes. Our small-world models, called SWANNs, provide several intriguing benefits: they facilitate data (gradient) flow within the network, enable feature-map reuse by adding long-range connections and accommodate various network architectures/datasets. Compared to densely connected networks (e.g., DenseNets), SWANNs require a substantially fewer number of training parameters while maintaining a similar level of classification accuracy. We evaluate our networks on various DL model architectures and image classification datasets, namely, MNIST, CIFAR10, CIFAR100, and ImageNet. Our experiments demonstrate an average of ≈2.1× improvement in convergence speed to the desired accuracy.

Automated Classification of Breast Cancer Cells Using High-Throughput Holographic Cytometry

Holographic cytometry is an ultra-high throughput quantitative phase imaging modality that is capable of extracting subcellular information from millions of cells flowing through parallel microfluidic channels. In this study, we present our findings on the application of holographic cytometry to distinguishing carcinogen-exposed cells from normal cells and cancer cells. This has potential application for environmental monitoring and cancer detection by analysis of cytology samples acquired via brushing or fine needle aspiration. By leveraging the vast amount of cell imaging data, we are able to build single-cell-analysis-based biophysical phenotype profiles on the examined cell lines. Multiple physical characteristics of these cells show observable distinct traits between the three cell types. Logistic regression analysis provides insight on which traits are more useful for classification. Additionally, we demonstrate that deep learning is a powerful tool that can potentially identify phenotypic differences from reconstructed single-cell images. The high classification accuracy levels show the platform’s potential in being developed into a diagnostic tool for abnormal cell screening.

Active Learning Framework for Time-Series Classification of Vibration and Industrial Process Data

Data-based models built using machine learning solutions are becoming more prominent in the condition monitoring, maintenance, and prognostics fields. The capacity to build these models using a machine learning approach depends largely in the quality of the data. Of particular importance is the availability of labelled data, which describes the conditions that are intended to be identified. However, properly labelled data that is useful in many machine learning strategies is a scare resource. Furthermore, producing high-quality labelled data is expensive, time-consuming and a lot of times inaccurate given the uncertainty surrounding the labeling process and the annotators. &#x0D; Active Learning (AL) has emerged as a semi-supervised approach that enables cost and time reductions of the labeling process. This approach has had a delayed adoption for time series classification given the difficulty to extract and present the time series information in such a way that it is easy to understand for the human annotator who incorporates the labels. This difficulty arises from the large dimensionality that many of these time series possess. This challenge is exacerbated by the cold-start problem, where the initial labelled dataset used in typical AL frameworks may not exist. Thus, the initial set of labels to be allocated to the time series samples is not available. This last challenge is particularly common on many condition monitoring applications where data samples of specific faults or problems does not exist.&#x0D; In this article, we present an AL framework to be used in the classification of time series from industrial process data, in particular vibration waveforms originated from condition monitoring applications. In this framework, we deal with the absence of labels to train an initial classification model by introducing a pre-clustering step. This step uses an unsupervised clustering algorithm to identify the number of labels and selects the points with a stronger group belonging as initial samples to be labelled in the active learning step. Furthermore, this framework presents two approaches to present the information to the annotator that can be via time-series imaging and automatic extraction of statistical features. Our work is motivated by the interest to facilitate the effort required for labeling time-series waveforms, while maintaining a high level of accuracy and consistency on those labels. In addition, we study the number of time-series samples that require to be labelled to achieve different levels of classification accuracy, as well as their confidence intervals. These experiments are carried out using vibration signals from a well-known rolling element bearing dataset and typical process data from a production plant. &#x0D; An active learning framework that considers the conditions of the data commonly found in maintenance and condition monitoring applications while presenting the data in ways easy to interpret by human annotators can facilitate the generation reliable datasets. These datasets can, in turn, assist in the development of data-driven models that describe the many different processes that a machine undergoes.

Machine Learning and Medication Adherence: Scoping Review.

This is the first scoping review to focus broadly on the topics of machine learning and medication adherence. This review aims to categorize, summarize, and analyze literature focused on using machine learning for actions related to medication adherence. PubMed, Scopus, ACM Digital Library, IEEE, and Web of Science were searched to find works that meet the inclusion criteria. After full-text review, 43 works were included in the final analysis. Information of interest was systematically charted before inclusion in the final draft. Studies were placed into natural categories for additional analysis dependent upon the combination of actions related to medication adherence. The protocol for this scoping review was created using the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) guidelines. Publications focused on predicting medication adherence have uncovered 20 strong predictors that were significant in two or more studies. A total of 13 studies that predicted medication adherence used either self-reported questionnaires or pharmacy claims data to determine medication adherence status. In addition, 13 studies that predicted medication adherence did so using either logistic regression, artificial neural networks, random forest, or support vector machines. Of the 15 studies that predicted medication adherence, 6 reported predictor accuracy, the lowest of which was 77.6%. Of 13 monitoring systems, 12 determined medication administration using medication container sensors or sensors in consumer electronics, like smartwatches or smartphones. A total of 11 monitoring systems used logistic regression, artificial neural networks, support vector machines, or random forest algorithms to determine medication administration. The 4 systems that monitored inhaler administration reported a classification accuracy of 93.75% or higher. The 2 systems that monitored medication status in patients with Parkinson disease reported a classification accuracy of 78% or higher. A total of 3 studies monitored medication administration using only smartwatch sensors and reported a classification accuracy of 78.6% or higher. Two systems that provided context-aware medication reminders helped patients to achieve an adherence level of 92% or higher. Two conversational artificial intelligence reminder systems significantly improved adherence rates when compared against traditional reminder systems. Creation of systems that accurately predict medication adherence across multiple data sets may be possible due to predictors remaining strong across multiple studies. Higher quality measures of adherence should be adopted when possible so that prediction algorithms are based on accurate information. Currently, medication adherence can be predicted with a good level of accuracy, potentially allowing for the development of interventions aimed at preventing nonadherence. Monitoring systems that track inhaler use currently classify inhaler-related actions with an excellent level of accuracy, allowing for tracking of adherence and potentially proper inhaler technique. Systems that monitor medication states in patients with Parkinson disease can currently achieve a good level of classification accuracy and have the potential to inform medication therapy changes in the future. Medication administration monitoring systems that only use motion sensors in smartwatches can currently achieve a good level of classification accuracy but only when differentiating between a small number of possible activities. Context-aware reminder systems can help patients achieve high levels of medication adherence but are also intrusive, which may not be acceptable to users. Conversational artificial intelligence reminder systems can significantly improve adherence.

Classification Analysis of Student Graduation Timeliness Using Decision Tree and Naïve Bayes Methods

This study aims to determine the classification of student graduation timeliness by using the Decision Tree and Naive Bayes methods. This study uses a quantitative method, where the approach used is the classification of various attributes that affect the timeliness of student graduation. The independent variables in the classification are mostly called attributes; In this study, the attributes of school of origin, gender, area of origin, profession of parents, study program and Grade Point Average (GPA) were used. While the dependent variable or in the classification is usually called a label, in this study the label used as a decision attribute is the timeliness of student graduation. In this study, two methods were used, namely using the nave Bayes method and a decision tree (decision tree) to determine the classification of the timeliness of student graduation and to determine the level of classification accuracy. Based on the results of the analysis, it can be concluded that the classification using the nave Bayes method obtained 36 predicted data according to the actual data and 7 different predicted data from the actual data. Meanwhile, in the 42 decision tree method, the predicted data is in accordance with the actual data and there is only 1 predicted data that is different from the actual data. Decision Tree method has a lower classification error rate than the Naive Bayes method. The level of accuracy of prediction results using the Decision Tree method is higher than the Naive Bayes method.

Are Microcontrollers Ready for Deep Learning-Based Human Activity Recognition?

The last decade has seen exponential growth in the field of deep learning with deep learning on microcontrollers a new frontier for this research area. This paper presents a case study about machine learning on microcontrollers, with a focus on human activity recognition using accelerometer data. We build machine learning classifiers suitable for execution on modern microcontrollers and evaluate their performance. Specifically, we compare Random Forests (RF), a classical machine learning technique, with Convolutional Neural Networks (CNN), in terms of classification accuracy and inference speed. The results show that RF classifiers achieve similar levels of classification accuracy while being several times faster than a small custom CNN model designed for the task. The RF and the custom CNN are also several orders of magnitude faster than state-of-the-art deep learning models. On the one hand, these findings confirm the feasibility of using deep learning on modern microcontrollers. On the other hand, they cast doubt on whether deep learning is the best approach for this application, especially if high inference speed and, thus, low energy consumption is the key objective.

Short and Long Time Bloodstains Age Determination by Colorimetric Analysis: A Pilot Study.

Bloodstains found at crime scenes represent a crucial source of information for investigative purposes. However, in forensic practice, no technique is currently used to estimate the time from deposition of bloodstains. This preliminary study focuses on the age estimation of bloodstains by exploiting the color variations over time due to the oxidation of the blood. For this purpose, we used a colorimetric methodology in order to easily obtain objective, univocal and reproducible results. We developed two bloodstain age prediction algorithms: a short-term and a long-term useful model for the first 24h and 60 days, respectively. Both models showed high levels of classification accuracy, particularly for the long-term model. Although a small-scale study, these results improve the potential application of colorimetric analysis in the time-line reconstruction of violent criminal events.

Two Stages Outlier Removal as Pre-processing Digitizer Data on Fine Motor Skills (FMS) Classification Using Covariance Estimator and Isolation Forest

The increase of the classification accuracy level has become an important problem in machine learning especially in diverse data-set that contain the outlier data. In the data stream or the data from sensor readings thatproduce large data, it allows a lot of noise to occur. It makes the performance of the machine learning model is disrupted or even decreased. Therefore, clean data from noise is needed to obtain good accuracy and to improve the performance of the machine learning model. This research proposes a two-stages for detecting and removing outlier data by using the covariance estimator and isolation forest methods as pre-processing in the classification process to determine fine motor skill (FMS). The dataset was generated from the process of recording data directly during cursive writing by using a digitizer. The data included the relative position of the stylus on the digitizer board. x position, y position, z position, and pressure values are then used as features in the classification process. In the process of observation and recording, the generated data was very huge so some of them produce the outlier data. From the experimental results that have been implemented, the level of accuracy in the FMS classification process increases between 0.5-1% by using the Random Forest classifier after the detection and outlier removal by using covariance estimator and isolation forest. The highest accuracy rate achieves 98.05% compared to the accuracy without outlier removal, which is only about 97.3%.

Spectral signature analysis to determine mangrove species delineation structured by anthropogenic effects

In this study, we used the spectral information divergence (SID) and spectral angle mapper (SAM) algorithm classification to identify ~ 40,000 individual trees from 19 mangrove species; these data were collected from 7 sampling points spread across 40,288 ha of the Matang Mangrove Forest Reserve (MMFR) in the state of Perak, Malaysia. This information was examined and analysed to answer two questions: (1) can medium resolution of satellite image identify the individual mangrove species (2) how are the distributions of individual species within the mangrove zonation associated with the anthropogenic matrix? Species identification with spectral library derived from the in-situ measurements using the SID algorithm was compared with that derived from the Landsat 8 using the SAM algorithm; it was found that the two methods offered different but complementary information with different rates of accuracy. The observed levels of classification accuracy are at 84.95% and 85.21%, respectively for SID and SAM algorithm classification. The mangrove species distribution has a correlation with the anthropogenic activities, but the distribution occurred randomly without specific zones. All the 19 selected mangrove species are closely related with different correlation by the anthropogenic activities. The use of the SID and SAM algorithms may provide the most promising way of classification for improving the mangrove species identification, with medium resolution of satellite image. Recognising the characteristics of mangrove zonation may give a better understanding of mangrove species appearances and conservation. However, mangrove zonation will remain a daunting mapping task and is a challenge for the ecologists in the future.

Metadata based Classification Techniques for Knowledge Discovery from Facebook Multimedia Database

Classification is a parlance of Data Mining to genre data of different kinds in particular classes. As we observe, social media is an immense manifesto that allows billions of people share their thoughts, updates and multimedia information as status, photo, video, link, audio and graphics. Because of this flexibility cloud has enormous data. Most of the times, this data is much complicated to retrieve and to understand. And the data may contain lot of noise and at most the data will be incomplete. To make this complication easier, the data existed on the cloud has to be classified with labels which is viable through data mining Classification techniques. In the present work, we have considered Facebook dataset which holds meta data of cosmetic company’s Facebook page. 19 different Meta Data are used as main attributes. Out of those, Meta Data ‘Type’ is concentrated for Classification. Meta data ‘Type’ is classified into four different classes such as link, status, photo and video. We have used two favored Classifiers of Data Mining that are, Bayes Classifier and Decision Tree Classifier. Data Mining Classifiers contain several classification algorithms. Few algorithms from Bayes and Decision Tree have been chosen for the experiment and explained in detail in the present work. Percentage split method is used to split the dataset as training and testing data which helps in calculating the Accuracy level of Classification and to form confusion matrix. The Accuracy results, kappa statistics, root mean squared error, relative absolute error, root relative squared error and confusion matrix of all the algorithms are compared, studied and analyzed in depth to produce the best Classifier which can label the company’s Facebook data into appropriate classes thus Knowledge Discovery is the ultimate goal of this experiment.

Real-Time Classification of Hand Motions Using Electromyography Collected from Minimal Electrodes for Robotic Control

In recent years, advances in human robot interaction (HRI) has shown massive potential for universal control of robots. Among them, electromyography (EMG) signals generated by motions of muscles have been identified as an important and useful source. Powered by recently emerged machine learning algorithms, real-time classification has been proved applicable to control robots. However, collecting EMG signals with minimum number of electrodes for real-time classification and robotic control is still a challenge. In this paper, we demonstrate that twenty five robotic commands in a robotic arm can be controlled in real time by using the EMG signals collected from only two pairs of active surface electrodes on each forearm of human subjects. To achieve this task, a variety of tested ML models for this classification were tested. Among them, the Gaussian Naïve Bayes (GNB) achieved an accuracy of &gt;96%. This unprecedented level of classification accuracy of the EMG signals collected from the least number of active electrodes suggest that by combination of optimized electrode configuration and a suitable ML model, the capability of robotic control can be maximized.

Schizophrenia recognition based on the phase space dynamic of EEG signals and graphical features

Schizophrenia is a mental disorder that causes adverse effects on the mental capacity of a person, emotional inclinations, and quality of personal and social life. The official statistics reported that about 20 million people suffer from this severe mental illness worldwide. The manual screening of schizophrenic patients is tedious, time-consuming, costly, and prone to human error. Thus, it is necessary to provide a fully automatic, fairly accurate, and reasonably inexpensive system to diagnose schizophrenia patients. Electroencephalography (EEG) is commonly employed to evaluate and detect the brain's functions and disorders. The purpose of this study is to introduce, apply, and examine a novel framework to automatically diagnose schizophrenia disorders. This procedure is performed by using the phase space dynamic (PSD) of EEG signals. The two-dimensional PSD of EEG signals is first plotted on Cartesian space. Then, fifteen graphical features are extracted to evaluate the chaotic behavior of PSD based on healthy and schizophrenic subjects. Also, a sizeable number of remarkable features and optimum channels are obtained by the forward selection algorithm (FSA). Finally, eight different classifiers are tested for schizophrenia detection. In this case, the K-nearest neighbor (KNN) and generalized regression neural network (GRNN) showed better performance than the others. As a result, using a 10-fold cross-validation strategy, the KNN classifier with City-block distance reached the level of the average classification accuracy (ACC) of 94.80%, the sensitivity (SEN) of 94.30%, and the specificity (SPE) of 95.20%. The findings of the study confirmed that the PSD shape of the Cz channel for schizophrenia groups is more regular than the healthy ones. It can be applied as a biomarker for the medical team to diagnose schizophrenia disorder. It was found that the frontal and parietal lobes reflect the effects of schizophrenia disorder better than the other lobes. Consequently, the proposed framework contributes to realizing a real-time, easily accessible, and fairly inexpensive method in clinics and hospitals to quickly detect schizophrenia disorder.

Habitat classification using convolutional neural networks and multitemporal multispectral aerial imagery

The monitoring of threatened habitats is a key objective of European environmental policies. Due to the high cost of current field-based habitat mapping techniques, there is keen interest in proposing solutions that can reduce cost through increased levels of automation. Our study aims to propose a habitat mapping solution that benefits both from the merits of convolutional neural networks (CNNs) for image classification tasks, as well as from the high spatial, spectral, and multitemporal unmanned aerial vehicle image data, which shows great potential for accurate vegetation classification. The proposed CNN-based method uses multitemporal multispectral aerial imagery for the classification of threatened coastal habitats in the Maharees (Ireland) and shows a high level of classification accuracy.

Fuzzy-Based Multi-Crop Classification Using High Resolution UAV Imagery

Accurate information regarding crop classification is important to estimation crop yield. It is used to depict the relationship between exponentially growing world population and food demand. The purpose of this research is to recognize multiple crops in a single UAV-based image. The task itself is chaotic as every crop exhibits similar hue, color and other plant characteristics. In this paper, an Adaptive Neuro-Fuzzy Inference System (ANFIS) is proposed to classify 17 different crops based on their high spatial and temporal signatures of normalized difference vegetation index (NDVI) values acquired through multispectral sensor onboard a quadrotor. The multispectral images were classified into two classes (soil and crop) and NDVI signatures for each crop were extracted from images. Detailed dataset was prepared as a timeline through sampling, covering almost all phenological phases of the crops. The NDVI dataset was passed through ANFIS to classify NDVI vectors. ANFIS had only one output variable: the crop type that was formulated from 8 input variables. ANFIS used 2 membership functions for one input variable and formulated 256 fuzzy rules for the classification. The results show a high level of classification accuracy.