Image Classification Techniques Research Articles

Critical Review of Processing and Classification Techniques for Images and Spectra in Microplastic Research.

Microplastic research is a rapidly developing field, with urgent needs for high throughput and automated analysis techniques. We conducted a review covering image analysis from optical microscopy, scanning electron microscopy, fluorescence microscopy, and spectral analysis from Fourier transform infrared (FT-IR) spectroscopy, Raman spectroscopy, pyrolysis gas-chromatography mass-spectrometry, and energy dispersive X-ray spectroscopy. These techniques were commonly used to collect, process, and interpret data from microplastic samples. This review outlined and critiques current approaches for analysis steps in image processing (color, thresholding, particle quantification), spectral processing (background and baseline subtraction, smoothing and noise reduction, data transformation), image classification (reference libraries, morphology, color, and fluorescence intensity), and spectral classification (reference libraries, matching procedures, and best practices for developing in-house reference tools). We highlighted opportunities to advance microplastic data analysis and interpretation by (i) quantifying colors, shapes, sizes, and surface topologies with image analysis software, (ii) identifying threshold values of particle characteristics in images that distinguish plastic particles from other particles, (iii) advancing spectral processing and classification routines, (iv) creating and sharing robust spectral libraries, (v) conducting double blind and negative controls, (vi) sharing raw data and analysis code, and (vii) leveraging readily available data to develop machine learning classification models. We identified analytical needs that we could fill and developed supplementary information for a reference library of plastic images and spectra, a tutorial for basic image analysis, and a code to download images from peer reviewed literature. Our major findings were that research on microplastics was progressing toward the use of multiple analytical methods and increasingly incorporating chemical classification. We suggest that new and repurposed methods need to be developed for high throughput screening using a diversity of approaches and highlight machine learning as one potential avenue toward this capability.

Prediction and Classification into Benign and Malignant using the Clinical Testing Features

Breast Cancer is the most often identified cancer among women and a major reason for the increased mortality rate among women. As the diagnosis of this disease manually takes long hours and the lesser availability of systems, there is a need to develop the automatic diagnosis system for early detection of cancer. The advanced engineering of natural image classification techniques and Artificial Intelligence methods has largely been used for the breast-image classification task. Data mining techniques contribute a lot to the development of such a system, Classification, and data mining methods are an effective way to classify data. For the classification of benign and malignant tumors, we have used classification techniques of machine learning in which the machine learns from the past data and can predict the category of new input. This study is a relative study on the implementation of models using Support Vector Machine (SVM), and Naïve Bayes on Breast cancer Wisconsin (Original) Data Set. With respect to the results of accuracy, precision, sensitivity, specificity, error rate, and f1 score, the efficiency of each algorithm is measured and compared. Our experiments have shown that SVM is the best for predictive analysis with an accuracy of 99.28% and naïve Bayes with an accuracy of 98.56%. It is inferred from this study that SVM is the well-suited algorithm for prediction.

Binary cross entropy with deep learning technique for Image classification

Binary cross entropy with deep learning technique for Image classification

Ultra-High-Resolution Mapping of Posidonia oceanica (L.) Delile Meadows through Acoustic, Optical Data and Object-based Image Classification

In this study, we present a framework for seagrass habitat mapping in shallow (5–50 m) and very shallow water (0–5 m) by combining acoustic, optical data and Object-based Image classification. The combination of satellite multispectral images-acquired from 2017 to 2019, together with Unmanned Aerial Vehicle (UAV) photomosaic maps, high-resolution multibeam bathymetry/backscatter and underwater photogrammetry data, provided insights on the short-term characterization and distribution of Posidonia oceanica (L.) Delile, 1813 meadows in the Calabrian Tyrrhenian Sea. We used a supervised Object-based Image Analysis (OBIA) processing and classification technique to create a high-resolution thematic distribution map of P. oceanica meadows from multibeam bathymetry, backscatter data, drone photogrammetry and multispectral images that can be used as a model for classification of marine and coastal areas. As a part of this work, within the SIC CARLIT project, a field application was carried out in a Site of Community Importance (SCI) on Cirella Island in Calabria (Italy); different multiscale mapping techniques have been performed and integrated: the optical and acoustic data were processed and classified by different OBIA algorithms, i.e., k-Nearest Neighbors’ algorithm (k-NN), Random Tree algorithm (RT) and Decision Tree algorithm (DT). These acoustic and optical data combinations were shown to be a reliable tool to obtain high-resolution thematic maps for the preliminary characterization of seagrass habitats. These thematic maps can be used for time-lapse comparisons aimed to quantify changes in seabed coverage, such as those caused by anthropogenic impacts (e.g., trawl fishing activities and boat anchoring) to assess the blue carbon sinks and might be useful for future seagrass habitats conservation strategies.

APPLICATION OF REMOTE SENSING AND GOOGLE EARTH ENGINE FOR MONITORING ENVIRONMENTAL DEGRADATION IN THE NILGIRI BIOSPHERE RESERVE AND ITS ECOSYSTEM OF WESTERN GHATS, INDIA

Abstract. Biosphere Reserves are archetypal parts of natural and cultural landscapes encompassing over large area of different ecosystem, it represents bio-geographic zones of an region. Globally, the areas of biosphere reserve is shrinking and exploiting due to the extreme climatic condition, natural calamities and anthropogenic activities, which leads to environmental and land degradation. In this paper Nilgiri Biosphere Reserve (NBSR) area has been selected and it represents a biodiversity-rich ecosystem in the Western Ghats and includes two of the ten biogeographical provinces of India. Amongst the most insubstantial ecosystems in the world, the Nilgiri Biosphere Reserve is bearing the substance of climate change evident in increasingly unpredictable rainfall and higher temperatures during recent years. The region was mostly unscathed till two centuries ago, but has witnessed large-scale destruction ever since. In this scenario, a need of application of remote sensing and advance machine learning techniques to monitor environmental degradation and its ecosystem in NBSR is more essential. The objective of the present study is to develop satellite image classification techniques that can reliably to map forest cover and land use, and provide the basis for long-term monitoring. Advanced image classification techniques on the cloud-based platform Google Earth Engine (GEE) for mapping vegetation and land use types, and analyse their spatial distributions. To restore degraded ecosystems to their natural conditions through proper management and conservation practices. In order to understand the nature of environmental degradation and its ecosystem in Nilgiri Biosphere Reserve; following thematic criteria’s were grouped in to four major indicators such as Terrain Indicator (TI), Environmental Indicator (EI), Hydro-Meteorological Indicator (HMI) and Socio-Economic Indicator (SEI). The utilisation of remote sensing product of huge datasets and various data product in analysis and advanced machine learning algorithm through Google earth engine are indispensable. After extraction of all the thematic layers by using multi criteria decision and fuzzy linear member based weight and ranks were assigned and overlay in GIS environment at a common pixel size of 30 m. Based on the analysis the resultant layer has been classified into five environmental degraded classes i.e., very high, high, moderate, slight and no degradation. This study is help to identify the degradation and long term monitoring and suggest the appropriate conservation, management and policies, it is a time to implement and protect the Nilgiri biosphere reserves without hindering present stage of natural environment in a sustainable manner.

LAND USE AND LAND COVER MAPPING – A CASE STUDY OF AHMEDABAD DISTRICT

Abstract. Land cover mapping using remote-sensing imagery has attracted significant attention in recent years. Classification of land use and land cover is an advantage of remote sensing technology which provides all information about land surface. Numerous studies have investigated land cover classification using different broad array of sensors, resolution, feature selection, classifiers, Classification Techniques and other features of interest from over the past decade. One, Pixel based image classification technique is widely used in the world which works on their per pixel spectral reflectance. Classification algorithms such as parallelepiped, minimum distance, maximum likelihood, Mahalanobis distance are some of the classification algorithms used in this technique. Other, Object based image classification is one of the most adapted land cover classification technique in recent time which also considers other parameters such as shape, colour, smoothness, compactness etc. apart from the spectral reflectance of single pixel.At present, there is a possibility of getting the more accurate information about the land cover classification by using latest technology, recent and relevant algorithms according to our study. In this study a combination of pixel-by-pixel image classification and object based image classification is done using different platforms like ArcGIS and e-cognition, respectively. The aim of the study is to analyze LULC pattern using satellite imagery and GIS for the Ahmedabad district in the state of Gujarat, India using a LISS-IV imagery acquired from January to April, 2017. The over-all accuracy of the classified map is 84.48% with Producer’s and User’s accuracy as 89.26% and 84.47% respectively. Kappa statistics for the classified map are calculated as 0.84. This classified map at 1:10,000 scale generated using recent available high resolution space borne data is a valuable input for various research studies over the study area and also provide useful information to town planners and civic authorities. The developed technique can be replicated for generating such LULC maps for other study areas as well.

A lightweight CNN for Diabetic Retinopathy classification from fundus images

Diabetic Retinopathy (DR) is a complication of diabetes mellitus that damages blood vessel networks in the retina. This is a serious vision-threatening issue in most diabetic subjects. The DR diagnosis by color fundus images involves skilled clinicians to recognize the presence of lesions in the image that can be used to detect the disease properly, making it a time-consuming process. Effective automated detection of DR is a challenging task. The feature extraction plays an excellent role in effective automated disease detection. Convolutional Neural Networks (CNN) have superior image classification efficiency in the present scenario compared to earlier handcrafted feature-based image classification techniques. This work presents a novel CNN model to extract features from retinal fundus images for better classification performance. The CNN output features are used as input for different machine learning classifiers in the suggested system. The model is evaluated through various classifiers (Support Vector Machine, AdaBoost, Naive Bayes, Random Forest, and J48) by using images from generic IDRiD, MESSIDOR, and KAGGLE datasets. The efficacy of the classifier is evaluated by comparing the specificity, precision, recall, False Positive Rate (FPR), Kappa-score, and accuracy values for each classifier. The evaluation results indicate that the proposed feature extraction technique along with the J48 classifier outperforms all the other classifiers for MESSIDOR, IDRiD, and KAGGLE datasets with an average accuracy of 99.89% for binary classification and 99.59% for multiclass classification. Furthermore, for the J48 classifier, the average Kappa-score (K-score) is 0.994 for binary classification and 0.994 for multi-class classification.

Season, Classifier, and Spatial Resolution Impact Honey Mesquite and Yellow Bluestem Detection using an Unmanned Aerial System

In Texas, mesquite and yellow-bluestem invasions are widespread. Identifying and monitoring juvenile and adult plants using high-resolution imagery from airborne sensors while they colonize new areas across the landscape can help land managers prioritize locations for treatment and eradication. In this study, we evaluated how data collection design using an unmanned aerial system (UAS) can affect plant detection and mapping. We used a Phantom 3 Professional unmanned aerial vehicle with a Parrot Sequoia multispectral camera for detecting and mapping native honey mesquite (Prosopis glandulosa) and non-native yellow bluestem (Bothriochloa ischaemum) at a rangeland site in northwest Texas. Flights were conducted seasonally during the period from summer 2017 to fall 2018 to test the seasonal impact of detecting plant species. Flights were conducted at altitudes of 30, 60, and 100 m, and four image classification techniques were tested to determine their viability of detecting distinct plant species. Results suggest that flights at 100-m aircraft altitude during the spring season are more effective (>80% user accuracies) for mapping mesquite canopies based on reflectance values and image segmentation information. Yellow bluestem mapping accuracies were low (< 20% user accuracies). Lower spatial resolution (100-m altitude flights, 12-cm pixel resolution) provided less noise and more generalization capabilities for the image classification methods. Overall, random forests and Support Vector Machine classification algorithms outperformed probability-based image classifiers. Land owners and rangeland ecologists using their own UAS in rangeland management can use this information to plan their data collection campaigns before the application of chemical treatments or manual eradication.

Use of multi-temporal satellite data for monitoring pool surface areas occurring in non-perennial rivers in semi-arid environments of the Western Cape, South Africa

Hydrological monitoring networks do not adequately cover non-perennial rivers in most parts of Africa partly because river flows do not frequently occur on these rivers. However, in the arid and semi-arid regions flows occurring on these rivers are an important resource that have to be managed judiciously. The lack of data on magnitudes, timing and frequency of non-perennial river flows constrain sustainable management of their water resources. The availability of moderate resolution satellite data (e.g. Sentinel-2 Multispectral Instrument and Landsat 8 Operational Land Imager) offers an opportunity to monitor the surface water availability specifically changes in surface areas covered with water in pools along these rivers. In this study, we investigated the use of remote sensing data to detect and monitor changes of water surface areas of pools along three non-perennial rivers and one perennial river in the Western Cape, South Africa, from 2016 to 2017, using Landsat 8 and Sentinel-2 datasets. We assessed a range of image pre-processing and image classification techniques to determine the most suitable method for surface water identification and pool area estimation along the selected rivers. The following indices were used; Normalised Difference Water Index (NDWI), Modified NDWI (MNDWI), Normalised Difference Vegetation Index (NDVI), Automated Water Extraction Index for shadowed (AWEIsh), non-shadowed regions (AWEInsh), and the Multi-Band Water Index (MBWI). The NDWI computed from Sentinel-2 TOA reflectance datasets was found to be the most suitable index for mapping pools, with an overall accuracy ranging between 60 and 86%. This index was used successfully to establish the variations of surface areas of pools covered with water along the Breede, Nuwejaars, Tankwa and Touws Rivers during the 2016–2017 period. The temporal variations of pool areas could be explained as expected by changes of rainfall and potential evapotranspiration. Overall, the study demonstrated the feasibility of using moderate resolution datasets to monitor pool changes of areas covered with water in non-perennial rivers, which contributes to improved management of water resources particularly in arid and semi-arid environments.

Detection of Breast Lesion Using Improved GLCM Feature Based Extraction in Mammogram Images

A classic mammogram image processing method commonly illustrates of mammogram image pre-processing, segmentation, feature extraction and classification techniques. Feature extraction is a significant measure in Computer Assisted Diagnosis (CAD) of Breast cancer abnormalities. Feature Extraction is the method of shrinking the dimension of image data by finding an essential detail as of the segmented image. The visual substance of a segmented image portions can be confined using this procedure. Feature extraction is one of the Image retrieval methods for achieving higher accuracy. In this paper an Improved Gray Level Co-occurrence Matrix (GLCM) Feature based Extraction with shape and Tamura features are discussed. Tamura features are extracted from Optimized Kernel Fuzzy Clustering Algorithm (OKFCA) which is one of the segmentation methods. This paper proposed Improved GLCM Feature Based Extraction algorithm to obtain the trained features which will be used for further processing in future. Experimental results of 322 images were evaluated from Mammographic Image Analysis Society (MIAS) database and used to evaluate the feature extraction process.

Automated optimized classification techniques for magnetic resonance brain images

This paper presents automatic tumor detection and classification approaches for brain magnetic resonance images (MRI). These approaches are based on hybrid-optimized classification techniques and classify brain MRI to healthy, benign or malignant. The proposed system implements three-optimization techniques combined with Artificial Neural Network (ANN). Multi-Verse Optimizer (MVO), Moth-Flame Optimizer (MFO) and Salp Swarm Algorithm (SSA) are used and compared to examine how these techniques could be successfully employed to enhance the classification accuracy via selecting the optimal parameters of ANN. The proposed techniques are applied to the Harvard database and BRATS challenge dataset to evaluate the performance via Receiver Operation Characteristics (ROC) analysis. The approaches are tested against geometric transformations such as scaling, rotation and warping to show how much the proposed system resists these transformations. Experimentally, the proposed algorithms achieve the highest classification accuracy as compared to the other published ones. Also, the MVO-ANN algorithm outperforms the other proposed algorithms.

Observing land-use/land-cover dynamics in semi-arid environment: Evidence from Damaturu Town and its surrounding lands, Yobe State, Nigeria

The fragile Sudano-Sahelian ecological zone of Nigeria has been classified as a hotspot of land cover change (LCC) that has been suffering from serious anthropogenic and biophysical stresses. Damaturu, being the fastest growing town situated in the region happened to be a victim of this negative development. The purpose of this study is to remotely observe and assess the prevailing land-use/land-cover (LULC) dynamics of Damaturu town and its delicate surrounding lands from the year 1987-2017 study periods. To achieve this, a supervised image classification technique with Maximum Likelihood Classifier (MLC) algorithm was used in ERDAS Imagine version 15 software to classify the three epochs multi-temporal and multi-spectral Landsat imageries (TM 1987, ETM+7 2000 and OLI 2017). The classified LULC maps and their resulting statistics were then used to assess the spatio-temporal aspects of the observed changes by placing the results within the wider context of previous related literature and evidences. Findings revealed that the built-up area has been expanding since 1987 with an annual change rate of 4.5% between 1987-2000, and 5.3% during 2000-2017 respectively. The growth of the town is being accompanied by massive farmlands expansion and vegetal cover (trees and shrubs) lost making the surrounding arable lands seriously disturbed. Thus, if the observed trends continue, the entire studied region will be subjected to severe environmental hazard such as desertification. Overall, the study provides valuable information required for sustainable environmental management.

Effects of land use and land cover change on soil erosion potential in Berhe district: a case study of Legedadi watershed, Ethiopia

ABSTRACT This research was done in Legedadi watershed since the area was highly susceptible to soil erosion problem that could be aggravated by heavy rainfall, and steep slope. The aim of the present study was to examine the impact of LULC change on soil erosion potential in Legedadi watershed, and its implication on sedimentation problems. Multi-temporal Landsat imagery of TM (1985), TM (1997) and ETM+ (2013) were deployed to categorize a LULC map of the study watershed. Soil erosion risk was calculated by using GIS and RUSLE. Digital image processing and classification technique have been done using ERDAS Imagine 9. 2. Results indicate that there was a continuous and active LULC change observed in the watershed for the last 28 years. The settlement area and cultivated land were increased by 14.34% and 18.3%, whereas grazing land and bare land classes were reduced continuously by 25.74% and 18.47% between (1985–2103) respectively. The estimated annual soil losses were 0.0-330 t ha−1year−1in 2013 whereas in 1997 its range was 0-257.1 t ha−1year−1. A mean annual soil loss in the watershed was 54.19 t ha−1year−1 in 1997and 66.21 t ha−1year−1 in 2013, respectively. The average annual soil loss from each sub-watershed ranges from 34.57–89.19 t ha−1year−1. This calls for the coordinated action of the local communities and government in ensuring sustainable natural resource conservation systems considering the Climate Resilient Green Economy strategy of Ethiopia.

RETRACTED ARTICLE: Pre-processing on remotely sensed data with unsupervised classification analysis

The use of remote sensing and geographic information system (GIS) technologies grow drastically in recent years by ecologists around the globe. At present, using sophisticated sensors, there is a massive challenge in handling the high dense remotely captured information with spatial, spectral, temporal, and radiometric resolutions. This article addresses how to handle such large volume remotely sensed data using R programming with the aid of RStudio. We aim broadly two categories, such as image preprocessing and classification techniques on remotely sensed data. Image preprocessing methods such as false-color composite, pan-sharpening, single event upset error mitigation, and a view on the hyper spectral image. The other category comes with a focus on landscape spatial features analysis and unsupervised classification to analyze vegetation land. The CLARA (Clustering LARge Application) algorithm is used in this study which exploits k-medoids approaches for the unsupervised classification. Also, for results comparison, different vegetation indices such as normalized difference water index (NDWI), modified NDWI (MNDWI), and soil adjusted vegetation index are used for vegetation (land) analysis. Also, for the unsupervised classification, the Silhouette index is used to compare the clustering algorithms.

ECG-Derived Heart Rate Variability Interpolation and 1-D Convolutional Neural Networks for Detecting Sleep Apnea.

Feature extraction from ECG-derived heart rate variability signal has shown to be useful in classifying sleep apnea. In earlier works, time-domain features, frequency-domain features, and a combination of the two have been used with classifiers such as logistic regression and support vector machines. However, more recently, deep learning techniques have outperformed these conventional feature engineering and classification techniques in various applications. This work explores the use of convolutional neural networks (CNN) for detecting sleep apnea segments. CNN is an image classification technique that has shown robust performance in various signal classification applications. In this work, we use it to classify one-dimensional heart rate variability signal, thereby utilizing a one-dimensional CNN (1-D CNN). The proposed technique resizes the raw heart rate variability data to a common dimension using cubic interpolation and uses it as a direct input to the 1-D CNN, without the need for feature extraction and selection. The performance of the method is evaluated on a dataset of 70 overnight ECG recordings, with 35 recordings used for training the model and 35 for testing. The proposed method achieves an accuracy of 88.23% (AUC=0.9453) in detecting sleep apnea epochs, outperforming several baseline techniques.

Active Ensemble Deep Learning for Polarimetric Synthetic Aperture Radar Image Classification

Although deep learning has achieved great success in image classification tasks, its performance is subject to the quantity and quality of training samples. For classification of polarimetric synthetic aperture radar (PolSAR) images, it is nearly impossible to annotate the images from visual interpretation. Therefore, it is urgent for remote sensing scientists to develop new techniques for PolSAR image classification under the condition of very few training samples. In this letter, we take the advantage of active learning and propose active ensemble deep learning (AEDL) for PolSAR image classification. We first show that only 35\% of the predicted labels of a deep learning model's snapshots near its convergence were exactly the same. The disagreement between snapshots is non-negligible. From the perspective of multiview learning, the snapshots together serve as a good committee to evaluate the importance of unlabeled instances. Using the snapshots committee to give out the informativeness of unlabeled data, the proposed AEDL achieved better performance on two real PolSAR images compared with standard active learning strategies. It achieved the same classification accuracy with only 86% and 55% of the training samples compared with breaking ties active learning and random selection for the Flevoland dataset.

A Hybrid Grey Wolves Optimizer and Convolutional Neural Network for Pollen Grain Recognition

Melissopalynology, or pollen analysis of honey, is one of the areas that benefited greatly from image processing and analysis techniques, where melissopalynology is the science that studies the pollen contained in honey, using a microscopic examination. Nowadays, developing an automatic classification system for pollen identification presents a challenge. This article presents a metaheuristic for image segmentation to detect pollen grains in images. It is a swarm intelligence technique inspired from grey wolf hunting behavior in nature, centered around respecting the hierarchy of a pack. It was tested on a set of microscopic images of pollen grains. To evaluate pollen detection, we represented the detected pollen grains using two methods, grey-level based representations where we kept grey value of each pixel, and a binary mask-based technique, where a pixel could have only two values (1 or 0). Then, we used a convolutional neural network (CNN) technique for image classification to predict the specie of each pollen. The proposed system was tested on a set of microscopic images of pollen grains. The obtained performance measures of the system proved that the system is very successful.

Voice Command Recognition Using Biologically Inspired Time-Frequency Representation and Convolutional Neural Networks.

Voice command is an important interface between human and technology in healthcare, such as for hands-free control of surgical robots and in patient care technology. Voice command recognition can be cast as a speech classification task, where convolutional neural networks (CNNs) have demonstrated strong performance. CNN is originally an image classification technique and time-frequency representation of speech signals is the most commonly used image-like representation for CNNs. Various types of time-frequency representations are commonly used for this purpose. This work investigates the use of cochleagram, utilizing a gammatone filter which models the frequency selectivity of the human cochlea, as the time-frequency representation of voice commands and input for the CNN classifier. We also explore multi-view CNN as a technique for combining learning from different time-frequency representations. The proposed method is evaluated on a large dataset and shown to achieve high classification accuracy.

An efficient image classification technique for Synthetic Aperture Radar (SAR) images

An efficient image classification technique for Synthetic Aperture Radar (SAR) images

A survey of band selection techniques for hyperspectral image classification

Hyperspectral images usually contain hundreds of contiguous spectral bands, which can precisely discriminate the various spectrally similar classes. However, such high-dimensional data also contain highly correlated and irrelevant information, leading to the curse of dimensionality (also called the Hughes phenomenon). It is necessary to reduce these bands before further analysis, such as land cover classification and target detection. Band selection is an effective way to reduce the size of hyperspectral data and to overcome the curse of the dimensionality problem in ground object classification. Focusing on the classification task, this article provides an extensive and comprehensive survey on band selection techniques describing the categorisation of methods, methodology used, different searching approaches and various technical difficulties, as well as their performances. Our purpose is to highlight the progress attained in band selection techniques for hyperspectral image classification and to identify possible avenues for future work, in order to achieve better performance in real-time operation.