Satellite Image Recognition Research Articles

Satellite image recognition using ensemble neural networks and difference gradient positive-negative momentum

The modern machine learning theory finds application in many areas of human activity. One of the most dispersed tasks is pattern recognition on satellite images. It is difficult for a person to recognize a large number of images in a short time. It made the researchers develop the automation process, such as neural network engagement. The loss function minimization and ensemble learning raise the pattern recognition accuracy. We propose the robust difference gradient positive-negative momentum optimization algorithm that achieves the global minimum of the loss function with higher accuracy and fewer iterations than known analogs. Such an optimization algorithm contains the generalized average moving estimation approach and more effective learning rate control by additional parameters. The proposed optimizer has the regret-bound rate estimation, belonging to OT, and converges to the global minimum. However, the main problems in optimization theory are vanishing and blowing gradient values, where the standard gradient-based algorithms fail to achieve the required objective function value. The vanishing and blowing gradient problems meet in Rastrigin and Rosebrock test functions, where the proposed optimization algorithm attains the global extreme in the shortest number of iterations and has a more stable convergence process than state-of-the-art methods. Afterward, there are trained deep convolutional neural networks with different optimizers on satellite images from the University of California merced dataset containing 21 object classes, where the proposed algorithm gives the highest accuracy. There is a suggested ensemble-learning model consisting of 4 networks with different optimizers. The prediction results receive weight coefficients distributed according to the majority voting and ensemble neural network retrains with the higher pattern recognition accuracy. The suggested ensemble-learning model with the developed optimizer raised the accuracy by 1 %–4 % percentage points.

Improving deep learning performance by augmenting training data

Satellite image recognition is a crucial application of computer vision that has the po-tential to be applied in various fields such as disaster management, agriculture, and urban planning. The objective of this study is to determine the optimal amount of input data required and select the most effective methods of augmentation necessary for training a convolutional neural network (CNN) for satellite image recognition. To achieve this, we perform a series of experiments to investigate the effect of input data quantity on several performance metrics, including model accuracy, convergence, and generalization. Additionally, we explore the impact of various data augmentation techniques, such as rotation, scaling, and flipping, on model performance. The study suggests several strategies for identifying the saturation point and mitigating the effects of overtraining, in-cluding early stopping and dropout regularization. The findings from this study can significantly contribute to the development of more ef-ficient satellite recognition models. Furthermore, they can help improve the performance of existing models, in addition to providing guidance for future research. The study emphasizes the importance of carefully selecting input data and augmentation methods to achieve optimal performance in CNNs, which is fundamental in advancing the field of computer vision. In addition to the above, the study investigates the potential of transfer learning by pre-training the model on a related dataset and fine-tuning it on the satellite imagery dataset. This approach can reduce the amount of required data and training time and increase model performance. Overall, this study provides valuable insights into the optimal amount of input data and augmentation techniques for training CNNs for satellite image recognition, and its findings can guide future research in this area.

ПЕРСПЕКТИВЫ ПРИМЕНЕНИЯ МОДЕЛИ РАСПОЗНАВАНИЯ СПУТНИКОВЫХ СНИМКОВ ДЛЯ АНАЛИЗА МАКРОЭКОНОМИЧЕСКОЙ СИТУАЦИИ

The paper explores the possibilities and prospects for the implementation of hybrid cognitive models for predicting macroeconomic indicators based on fuzzy cognitive maps and neurofuzzy neural network models based on data on recognized natural fires on satellite images. In the course of the work, an analysis of existing satellite image recognition tools was carried out, problems in the field of satellite image recognition as a source of information were considered and analyzed, deep learning methods were compared, and existing image recognition methods were analyzed. Based on the results of the problem analysis, a deep neural network model was developed for recognizing natural fires on satellite images. Based on the obtained results of image classification, a fuzzy cognitive map for the analysis of the macroeconomic situation was constructed.

Temporal variance of the ice navigation conditions within the Northern Sea Route

Recognition of satellite images, composition of them and vectorization is used in AARI for ice charts production. There is discussed methodology and results of the ice charts processing by means of computer programs, which were elaborated by Dr. Tretyakov in Python. The paper demonstrates results of analysis of temporal variance of ice navigation circumstances within the buffer zone of the marine transport system from the Sabetta Port (the Yamal Peninsula, Russia) up to the Bering Strait. There are considered the variance for April and May from 1998 up to 2020. This intra-annual interval is the one with the heaviest ice circumstances for shipping. We used conditional length of various age and age and form gradations of the sea ice for the route as a whole, as an integral parameter for estimation of the navigation hardships of ice navigation. The conditional length of an ice age (thickness) diapason is result of multiplication of the diapason partial concentration at the length of the route leg with homogeneous ice characteristics. There were produced series of the conditional lengths for each ten-day periods during April and May. Then statistical homogeneity of the series was tested by various methods.

Alterations of the heaviest ice conditions along Sabetta-Bering Strait route

AARI has unique archive of ice charts, which were made by recognition of satellite images, their composition and vectorization. The paper is considered methodology and preliminary results of computer processing of vector ice charts from the archive. The archive covers the period since 1997. The charts handling is processed in ArcGIS. The article presents some results of ice navigation conditions analysis within the standard route “Sabetta Port – the Bering Strait” for April and May, that is temporal interval of the heaviest conditions of ice navigation. There have been calculated conditional length of various ice age/age and form gradations for the route as a whole. For a homogeneous ice zone of the route the conditional length of an ice age gradation means result of multiplication of the gradation partial ice concentration at the zone length. The values were calculated for various ice forms (floe dimensions) of the ice age categories. The research revealed essential enhancement of the ice navigation conditions for the period 1998-2020. Nevertheless even complete disappearance of the old ice from the route does not mean absence of nip danger.

Machine Learning In Problems Involved In Processing Satellite Images

The use of a machine (computer) vision system for remove probing of the Earth is considered. A survey of software and hardware methods used in computer vision systems in the processing of satellite images is presented. Methods of processing data with the use of a trained neural network are described. Examples of algorithmic processing of satellite images by means of artificial convolutional neural networks are presented. Methods of increasing the precision of recognition of satellite images are determined. Practical applications of convolutional neural networks onboard microsatellites used for remote probing of the Earth are presented.

Vessel movement velocity as the integral parameter of ice navigation conditions

The publication is considered methodology and preliminary results of computer processing of vector ice charts and simulation of ice navigation by empirical statistical model, which was designed in the AARI. The model is result of generalization of multi-year special ship ice observations, produced by the AARI scientists. The Institute has unique archive of ice charts, which were made by recognition of satellite images, their composition and vectorization. The archive covers the period since 1997. Average navigation velocity and total time expenditure for sailing along a route can be used as the objective indexes of the ice navigation conditions. In the article is presented detailed description of the technique for preparing data for numerical experiments with the empirical statistical model by applying the ice charts from the archive. The data preparing is processed in ArcGIS by means of specially designed computer programs. The numerical experiments simulate ice navigation of “Arctica” nuclear icebreaker. The comparison between the average navigation velocity and time expenditure for sailing along the route “Sabetta Port – Kara Gate Strait” at simulation of ice conditions in the first ten-day interval of May 1998 and the same temporal interval of 2019 indicates significant improve of the ice navigation conditions.

Possibilities of deep learning neural networks for satellite image recognition

The main problem solved in this project is the analysis of big data using a system of computer processing and recognition of satellite images, based on a deep neural network architecture. The goal of the project is to develop methodological, theoretical and practical aspects of building such systems in poorly formalized subject areas, as well as to study the possibilities and advantages of building predictive models for analyzing fresh water reserves and predicting the direction, speed and nature of the spread of large fires using such systems. and assessments of the economic impact of these natural disasters.

A New Feature Extraction Method Using Genetic Optimization for Texture Recognition of Satellite Images

In this paper, a new approach is proposed for texture feature extraction of satellite images using genetic algorithms. In this approach, the texture feature does not change by changing the angle of the texture. First, image texture is convolved with a trained mask, and then its energy is calculated. A regulatory mask has also been trained by a genetic algorithm using many examples of textured images. Feature extraction experiments are performed on a series of aerial images. Experimental results on both agricultural and urban images show the effectiveness of this feature for texture discrimination process of satellite images. Feature extraction for texture analysis is the key method that is used for texture recognition of satellite images. It is divided into four main categories: 1- structural methods; 2- model-based methods, 3- conversion methods, and 4- statistical methods (1-2). Texture of structural method is defined as micro-texture and macro-texture. The advantage of structural method describes a good symbolic from images and more for composition analysis is presented. This method is not suitable for normal textures, since great variability exists in micro-texture and macro-texture and cannot be identified to distinguish them. Texture analysis based on such models, fractal and Markov models are based on building an image, so that it can be used to describe the composition and texture. In this method, the image, as a possible model or as a combination of basic functions is described by a set of lines. Fractal model for modeling natural textures and sizes and analyzed texture and texture differentiation is useful, but this method is weak to choose the angle and describing the structure of the local image context is useful. Methods such as Fourier transform, Gabor and violet by using frequencies within the image, do texture analysis work, and each has its own weaknesses. Statistical methods indirectly describe the texture distribution and the relationship between the gray levels of the image handles. For images with natural texture, most statistical methods are used for feature extraction. In statistical methods, texture is specified as a probability distribution on the image space to form a random model or a set of statistical features. The most common features that are actually used for texture analysis, are based on harmony search and search pattern in the fabric. In the weakness in features from most of the above methods, there is a change in the angle of the texture that will change, and therefore the feature for classification in the texture of a different type than the angle changes does not stay constant, and it cannot be so strong in appearance (3). In continued histological classification of other methods (4-5), a method for extracting a feature here is useful to apply the texture, which is a measure of energy. Of previous methods for texture classification, the feature space is necessarily random search. In this method, a random search of the parameter space for masks (6) takes place before applying the texture of the image and calculating the energy, initially trained by type of texture, which results in a simple and effective classifier. We apply a convolution mask on the texture image and obtain the energy of each pixel at position (I, J) from image. In the proposed method, we set the parameters for the mask instead of the conventional randomized algorithms, and then the genetic algorithm is used. By using the genetic algorithm makes, this method is insensitive to the angle of the texture and the significance of the classifier of random algorithms will increase. The reminder paper is organized as follows: Section 2 introduces desirable texture features. In Section 3, genetic algorithm is used to optimize a mask to extract the optimal texture features, and experimental results are presented in Section 4.

Texture Segmentation and Edge Location of Regions in Multiband Images

Many texture classification schemes require an excessively large image area for texture analysis, use a large number of features to represent each texture or are computationally very demanding. This paper presents a segmentation method considering all image bands information that can be used on natural or synthetic texture, soft or rough motifs, it permits also distinction of different textures with few changes on the same type of patterns. The approach uses a proposed coefficient, CVE; to estimate region limits that can be used for very small to very large regions and permits correct real time texture bounder classification. The space positions among the pixels on the texel are considered. The bands or color informations are combined considering they mean value and its standard deviation by the new coefficient. This scheme is computationally very efficient and it is suitable for color texture, medical analysis or satellite image recognition. It can be used of all type of texture because the rules of what will be identified are completely given by the used and adapted to each situation.

Large-Scale Optimization-Based Classification Models in Medicine and Biology

We present novel optimization-based classification models that are general purpose and suitable for developing predictive rules for large heterogeneous biological and medical data sets. Our predictive model simultaneously incorporates (1) the ability to classify any number of distinct groups; (2) the ability to incorporate heterogeneous types of attributes as input; (3) a high-dimensional data transformation that eliminates noise and errors in biological data; (4) the ability to incorporate constraints to limit the rate of misclassification, and a reserved-judgment region that provides a safeguard against over-training (which tends to lead to high misclassification rates from the resulting predictive rule); and (5) successive multi-stage classification capability to handle data points placed in the reserved-judgment region. To illustrate the power and flexibility of the classification model and solution engine, and its multi-group prediction capability, application of the predictive model to a broad class of biological and medical problems is described. Applications include: the differential diagnosis of the type of erythemato-squamous diseases; predicting presence/absence of heart disease; genomic analysis and prediction of aberrant CpG island meythlation in human cancer; discriminant analysis of motility and morphology data in human lung carcinoma; prediction of ultrasonic cell disruption for drug delivery; identification of tumor shape and volume in treatment of sarcoma; discriminant analysis of biomarkers for prediction of early atherosclerois; fingerprinting of native and angiogenic microvascular networks for early diagnosis of diabetes, aging, macular degeneracy and tumor metastasis; prediction of protein localization sites; and pattern recognition of satellite images in classification of soil types. In all these applications, the predictive model yields correct classification rates ranging from 80 to 100%. This provides motivation for pursuing its use as a medical diagnostic, monitoring and decision-making tool.

Improving classification with latent variable models by sequential constraint optimization

In this paper we propose a method to use multiple generative models with latent variables for classification tasks. The standard approach to use generative models for classification is to train a separate model for each class. A novel data point is then classified by the model that attributes the highest probability. The algorithm we propose modifies the parameters of the models to improve the classification accuracy. Our approach is made computationally tractable by assuming that each of the models is deterministic, by which we mean that a data-point is associated to only a single latent state. The resulting algorithm is a variant of the support vector machine learning algorithm and in a limiting case the method is similar to the standard perceptron learning algorithm. We apply the method to two types of latent variable models. The first has a discrete latent state space and the second, principal component analysis, has a continuous latent state space. We compare the effectiveness of both approaches on a handwritten digit recognition problem and on a satellite image recognition problem.

Image recognition using analog-ART1 architecture augmented with moment-based feature extractor

In machine intelligence, pattern recognition plays a vital role. The capability of Grossberg's ART1 (Adaptive Resonance Theory) augmented with moment-based feature extractor to behave as a pattern recognizer/classifier of images both noisy and noise free has been investigated in this paper. The potential of ART1 based pattern recognizer to recognize patterns, monochrome and colour, noisy and noise free has been studied on two experimental problems. The first experiment concerns with monochrome imaging pertaining to recognition of satellite images, a problem discussed by Wang et al. (International Conference on Acoustics, Speech and Signal Processing, Vol. 3, San Francisco, CA, 1992, p. 145; Adaptive Fuzzy Systems and Control, Prentice-Hall, Englewood Cliffs, NJ, 1994). The second experiment concerns with colour images dealing with recognition of some sample test coloured patterns. The result of computer simulation is also presented.

IMAGE RECOGNITION USING SIMPLIFIED FUZZY ARTMAP AUGMENTED WITH A MOMENT BASED FEATURE EXTRACTOR

The capability of Kasuba's Simplified Fuzzy ARTMAP (SFAM) to behave as a Pattern Recognizer/Classifier of images both noisy and noise free has been investigated in this paper. This calls for augmenting the original Neuro–Fuzzy model with a modified moment-based RST invariant feature extractor. The potential of the SFAM based Pattern Recognizer to recognize patterns — monochrome and color, noisy and noise free — has been studied on two experimental problems. The first experiment which concerns monochrome images, pertains to recognition of satellite images, a problem discussed by Wang et al. The second experiment, which concerns color images, deals with the recognition of some sample test colored patterns. The results of the computer simulation have also been presented.