Introduction

Image Processing is one of the major fields of signal processing. In order to augment the worth of an image or to draw worthwhile information from an image, the images need to be manipulated using various image processing techniques (Jain 1989). With the help of digitization, an image is converted into a suitable form to store on a computer device. The fundamental requirements are that, image must be sampled and quantized. Once the image has been stored, image processing operations may be executed on the acquired image to get better and flawless information. Typical stages in image processing are acquisition and enhancement of an image. These techniques are used to carry out the restoration, segmentation (Zaitoun and Aqel 2015), object recognition, improvement in interpretability (Gajdhane and Deshpande 2014), better representation and description (Demirel et al. 2009) of an image [U4]. Image acquisition involves various pre-processing tasks such as histogram equalization, scaling, brightness control etc. These tasks are required for improvement of image data to suppress unwanted distortions for further processing. The image enhancement and restoration techniques primarily focus on refinements to control the appearance of an image (Demirel et al. 2009). The process of segregating an image into various sub-parts is known as image segmentation (Zaitoun and Aqel 2015). The image segmentation technique includes the partitioning of an image into sub-parts or objects and is a key step from image processing to image analysis. Recognition methods utilize the difference of grey values of an image (Lee et al. 1994). The image description provides the extraction of features to result in some quantifiable information of interest for segregating one class of objects from another class. The initial applications of image processing were majorly focused in the areas of news-paper and allied industries. With further advancements in technology, the application areas of image processing technology have increased to diversified fields such as medical imaging, Pattern recognition, robot vision and remote sensing imagery [U5].

The need for accurate, fast and cost-effective geospatial information provided by remote sensing technology is increasing day by day (Mohammadzadeh et al. 2009). The remote sensing technique contributes to take important decisions by providing the accurate information for a large area (Campbell and Wynne 2011). Many image processing and analysis techniques have been developed to aid the interpretation of remotely sensed images and to extract as much information as possible from the images. As the manual extraction and analysis techniques tend to be expensive with respect to efforts required, time consumed and quality. Therefore, there is a requirement for automatic image feature extraction technique to expedite the process, and thereby considerably reducing the cost, computation time and better interpretation of information from satellite images (Mohammadzadeh et al. 2009). Artificial intelligence (AI) may replicate the human intelligence model or some natural phenomena and is considered to be a part of machine learning (Alpaydin (2009). The conept of artificial intelligence has been introduced to develop human intelligence in machines (McCorduch 1979). AI is able to provide more precision with higher degree of accuracy. Enhanced capabilties for the spatial databases may also be developed with the help of AI (McKeown 1987). In AI, nature or bio inspired meta-heuristics algorithms models their behaviour upon the naturally occuring phenomena and are developed for optimization problems to provide sufficiently good results and makes sure that the computation or size are not increasing (Singh et al. 2017). AI based techniques are capable of improving the image analysis, to give precise information regarding geographic conditions and earth resources applications (Estes et al. 1986). Thus, the artificial intelligence techniques viz.- Genetic Algorithm, Particle Swarm Optimization, Cuckoo Search etc. may play a vital role to improve the quality and interpretation process of available satellite images. These techniques may be efficiently utilized to carry out various image processing operations such as image enhancement, segmentation, contrast variation, noise removal etc. [U6].

In satellite images due to huge data size, large time is consumed to interpret the correct information. Further, the quality of satellite images is affected by weather conditions. In order to reveal the finer information and to improve the visibility qualities of satellite images, there is a need for implementation of suitable artificial intelligence technique. This paper presents detailed comparisons of various swarm intelligence-based techniques such as ant colony optimization, artificial bee colony algorithm, particle swarm optimization, bat algorithm etc. with respect to classifier, utility, images used, observations including the advantages/disadvantages and comparisons [U7]. Further, applications, advantages and disadvantages of various swarm-based techniques in numerous areas of satellite images processing have been presented. The particle swarm optimization is one of the most widely used technique, further its application areas with future research scope have been discussed [U8]. The section 2 consists of literature review on implementation of swarm intelligence in satellite imagery. Section 3 consists of discussion drawn from section 2. The last section discusses the conclusion.

Remote sensing imagery

Remote Satellite Images are like reservoir of useful and interesting information. From these images, one can find out transformations of cities, crop classification and cultivation pattern, and damage assessment in case of natural calamities such as floods, fire, storms. Satellite images act like a rich repository of information related to agricultural activities. The focus on vegetation helps in understanding the crop growth from planting to harvest along with the observation of abnormalities and the change due to season progression (Senthilnath et al. 2008). Classification in remote sensing images is also a very crucial activity and is frequently carried out for obtaining the information related to land-use land-cover. Environmental changes at global, regional and local level can also be monitored closely with the help of land-use land-cover and the changes occurring over time (Turner et al. (1994) [U9]. Visible, Infrared and Water Vapor Imagery are the three important classifications of satellite images. Sunlight disseminated by objects suspended in the air or on Earth represents Visible Imagery. Infrared Imagery identifies the clouds by measuring the heat radiation using satellite sensors. Measurement of moisture in the upper atmosphere represents Water Vapor Image. Satellite image optimization with respect to registration, enhancement, classification and segmentation is achieved by various techniques such as discrete wavelet transformation, swarm intelligence, fuzzy logic, singular wave decomposition etc.

Swarm intelligence

Swarm Intelligence is an imperative concept in Artificial Intelligence with primary aim of performance optimization and robustness. Swarm intelligence is a bottom-up approach and behaves like a multi-agent system, where there are plentiful simple beings such as birds, fish, ants etc., and these beings work in full cooperation and competition among the individuals (Liu et al. 2008b) [U10]. Collective behavior emerged from social insects’ forms swarm intelligence in which social exchanges amongst the individual representative help in finding out the optimal solutions for NP-hard problems. Different sorts of swarm optimization which are being utilized for streamlining are Artificial Bee Colony, Particle Swarm Optimization, Firefly Algorithm, Ant Colony, Bat Algorithm and so on. Swarm intelligence has been successful in solving complex problems such as network routing, pattern recognition, travelling salesman problems, data clustering and is currently a hot research topic in artificial intelligence (Liu et al. 2008b). Classification and feature extraction research based on swarm intelligence indicates that the new clasiification and intelligence computation methods helps in avoiding the imapct on classified results generated by artificial fault or deviation, improves the clasification validity and humanity and in reforming the robustness of the algorithm to operation management (Dong and Xiang-bin 2008).[U11].

Literature review

Optimization of satellite images in terms of enhancement, segmentation, classification, clustering is accomplished by actualizing swarm intelligence methods such as Ant Colony, Particle Swarm Optimization, Cuckoo Search and so on. The following section covers the review of application of various swarm intelligence algorithms. The major advancements in the area of swarm intelligence techniques initiated from the year 2006 onwards; this study considers the period from 2006 to 2018 and is divided into four sub-sections. The strings such as swarm intelligence in remote sensing, artificial intelligence in remote sensing and nature inspired algorithms have been used with prime consideration of remote sensing images. [U12].

Analysis from 2006 to 2008

Das et al. (2006) presented a hybrid framework comprising Particle Swarm Intelligence (PSO) and Rough-Set theory for image clustering. Zhong et al. (2006) worked in the area of classification of remote sensed data by the application of an innovative approach using Unsupervised Artificial Immune Classifier. In the first step, the clustering centres were randomly carefully chosen from the input images and later the classification task was carried out. Omkar et al. (2007) implemented Ant Colony Optimization and Particle Swarm Optimization for satellite image classification problem of land cover mapping. Monteiro and Kosugi (2007) presented a feature selection algorithm for remote data by implementing Particle Swarm Optimization (PSO). The method utilized swarm implementation for optimizing desired performance criteria and the count of selected features simultaneously. Senthilnath et al. (2008) implemented Particle Swarm Optimization, Maximum Likelihood Classifier (MLC) and Ant Colony Optimization in the area of crop coverage classification using high resolution satellite images. Liu et al. (2008a) have proposed the use of Ant Colony Optimization (ACO) for improving the classification performance. It was detected that the ACO algorithm gives better accuracy and rule set as compared to See 5.0 Decision Tree process. Liu et al. (2008b) proposed a new method using Particle Swarm Optimization (PSO) for satellite image classification. PSO is capable of finding optimized cut points and has good convergence in the exploration process. Dong and Xiang-bin (2008) worked in the area of image classification of remote sensed data showing the application of Particle Swarm Optimization. The advantage of neighbourhood information is utilized by PSO and is also a robust approach and can be implemented for other kinds of image classification. The detailed comparison is summarized in Table 1.

Table 1 Comparative analysis from 2006 to 2008
Full size table

Analysis from 2009 to 2011

Mohammadzadeh et al. (2009) applied Particle Swarm Optimization (PSO) to a mean calculation system using fuzzy, for obtaining road mean value in each band. Maulik and Saha (2009) worked in the area of image classification and proposed a modified Differential Evolution (DE) using fuzzy clustering technique and also performed statistical significance tests for establishing the superiority. Chen and Leou (2009) have used Particle Swarm Optimization (PSO) for proposing a new IKONOS imagery fusion technique for Panchromatic (PAN) and Multispectral (MS) satellite images. The visual quality and correlation coefficients were better and greater than the other methods. Juneja et al. (2009) implemented and performed relative analysis of Artificial Neural Network (ANN), Rough-Set and Fuzzy-Rough classifier. Paoli et al. (2009) proposed unsupervised classification for hyperspectral images using swarm intelligence which optimized the Bhattacharyya statistical distance between classes and the log likelihood function.

Daamouche and Melgani (2009) implemented a novel classification scheme for hyperspectral images articulating wavelet optimization within Particle Swarm Optimization structure. Chang et al. (2009) worked in the area of hyperspectral image band selection by developing Greedy Modular Eigenspaces (GME) and a novel Parallel Particle Swarm Optimization (PPSO) was offered. The proposed technique improved the computational speed with the help of parallel computing techniques and better reliable solutions as compared to GME. Ding and Chen (2009) has proposed the use of Particle Swarm Optimization (PSO) for improving the Support Vector Machine (SVM) classifier performance for hyperspectral classification. It was observed that the SVM approach has superiority over other traditional classifiers.

Papa et al. (2010) implemented a hybrid Particle Swarm Optimization- Projections Onto Convex Sets (PSO-POCS) algorithm for remote sensing image restoration. Bedawi and Kamel (2010) proposed clustering algorithm using Particle Swarm Optimization (PSO) for segmentation of high resolution images and the output was matched with K-means. Linyi and Deren (2010) worked in the areas of image fuzzy classification by proposing the use of Particle Swarm Optimization (PSO) and evaluation was done with Genetic Algorithm (GA) and mean value method. Ari and Aksoy (2010) worked for estimation of likelihood of Gaussian Mixture Models by presenting a Particle Swarm Optimization (PSO) based method. New parameterization for random covariance matrices was also presented. Gupta et al. (2011) proposed an extension of Biogeography Based-Optimization (BBO) for image classification. The migration rate is determined by using Rank based fitness criteria. Very accurate land-cover features were extracted.

Halder et al. (2011) presented supervised and unsupervised Ant based classification and clustering methods for automatic generation of landuse map. Senthilnath et al. (2011b) applied a novel Glowworm Swarm Optimization clustering method in image classification for multispectral satellite images. Goel et al. (2011) presented an innovative Particle Swarm Optimization - Biogeography Based Optimization (PSO-BBO) hybrid approach for classification of multispectral remote images. The method is very efficient and accurate in terms of land cover feature extraction. Arora et al. (2011) has proposed the application of Particle Swarm Optimization (PSO) with morphological operators in the classification of urban features in the satellite images. Samadzadegan and Mahmoudi (2011) proposed the implementation of Firefly Algorithm (FA) and Particle Swarm Optimization (PSO) for band selection. The proposed method outperforms Genetic Algorithm (GA). Bedawi and Kamel (2011) used Particle Swarm Optimization (PSO) for classifying remote data over urban areas. The result shows the significance with high predictive accuracy. Zhang et al. (2011a) presented endmember extraction technique by employing Ant Colony Optimization (ACO) and compared the results to N-FINDR and VCA algorithms. Zhang et al. (2011b) proposed an endmember extraction method by means of Particle Swarm Optimization. Senthilnath et al. (2011a) used Discrete Particle Swarm Optimization in image registration and it turns out to be an efficient technique. Table 2 provides a detailed comparison.

Table 2 Comparative analysis from 2009 to 2011
Full size table

Analysis from 2012 to 2014

Senthilnath et al. (2012) proposed hierarchical clustering procedure by using Glowworm Swarm Optimization (GSO), Niche Particle Swarm Optimization (NPSO) and Mean Shift Clustering (MSC), and it was observed that GSO based approach was robust and more accurate. Wang et al. (2012) presented PSO based approach for post-processing the Sub-Pixel Mapping (SPM) results obtained with the help of Sub-Pixel/Pixel Spatial Attraction Model (SPSAM). Yamaguchi et al. (2012) applied Particle Swarm Optimization (PSO) to the problem of similar image search by using the concept of transfer learning. Banerjee et al. (2012) attempted to solve the image classification land-cover problem by implementing Artificial Bee Colony (ABC) and the comparison was made with other methods. Soliman et al. (2012) worked in the field of image classification by using Support Vector Machine (SVM) and PSO, and evaluation revealed that the usage of RBF kernel function had utmost exactness ratio as well as polynomial kernel. Gao et al. (2012) implemented Ant Colony Optimization (ACO), for endmember extraction, based on GPU and the results were evaluated. Yavari et al. (2013) presented modified Particle Swarm Optimization (PSO) in identifying the ideal terms for Rational function models (RFM). Bhandari et al. (2014a) employed Cuckoo Search (CS) and Wind Driven Optimization (WDO) along with the use of Kapur’s entropy for multilevel thresholding and revealed their efficiency and accuracy. Zarrinpanjeh et al. (2013) proposed ant-agent use in the updation of road map. Satisfactory results with respect to verification, detection and extraction of roads.

Senthilnath et al. (2013) used Genetic Algorithm (GA) and PSO for flood evaluation and river mapping and proved to be an accurate and reliable approach. Bhandari et al. (2014b) presented the application of Artificial Bee Colony algorithm with DWT-SVD for the enhancement in contrast. The proposed technique is better as compared to DCT-SVD, PSO, DWT-SVD, GHE and PSO’s modified versions. Ghosh et al. (2013) designed a supervised feature selection method with the help of Self-adaptive Differential Evolution (SADE). The techniques also used the method of feature ranking. (Zhang et al. (2013) have proposed methods for improving ACO algorithm for extraction of endmember. Bhandari et al. (2014c) presented Cuckoo Search (CS) and DWT-SVD for contrast enhancement and comparison was done in terms of Standard Deviation, MSE, PSNR and Mean. Ghamisi et al. (2014) used fractional-order Darwinian Particle Swarm Optimization for multilevel thresholding. Significant improvement with respect to CPU time and fitness value was observed. Xue et al. (2014) proposed HA-PSO-SVM for image classification which improved the classification performance as compared to other technique. Zhong et al. (2014) proposed adaptive Differential Evolution for endmember extraction. The technique extracted endmember with higher precision. The detailed comparison is presented in Table 3.

Table 3 Comparative analysis from 2012 to 2014
Full size table

Analysis from 2015 to 2018

Bhandari et al. (2015a) worked on finding the optimal multilevel thresholds by the use of modified Artificial Bee Colony (ABC) algorithm with various objective functions. The results are promising and minimized the computational time. Bhandari et al. (2015b) used Tsallis entropy function with Cuckoo Search (CS) algorithm for color image segmentation. The proposed technique selected very effectively and properly the threshold values. Agrawal and Bawane (2015) proposed new multiobjective Particle Swarm Optimization technique for determining different bands and the count of unseen layer nodes. Jayanth et al. (2015a) projected the use of Artificial Bee Colony (ABC) algorithm in satellite classification data and was compared with MLC, ANN and SVM. Ghamisi and Benediktsson (2015) proposed integrated Genetic Algorithm- Particle Swarm Organization (GA-PSO) for feature selection. It was confirmed that the approach automatically selected the most informative feature and was also tested for road detection. Senthilnath et al. (2015) used GA and NPSO for image registration and image clustering and the performance was compared to the conventional methods. Jayanth et al. (2015b) implemented Artificial Bee Colony algorithm for improving the performance of data classification. An enhancement of 5% was achieved in classification precision. Praveena and Singh (2014) presented the use of feed-forward neural networks classifier for image segmentation. Li et al. (2015) presented DPSO based flood inundation mapping- sub-pixel and comparison was done with other methods. Wang et al. (2015) proposed improved online dictionary learning involving Particle Swarm Optimization and the method had superior effect on noise suppression. Iounousse et al. (2015) developed an unsupervised technique on the bases of Probabilistic Neural Network and the accuracy results were compared with other methods. Upadhyay et al. (2010) used Artificial Neural Networks (ANN) for satellite image classification. Yang et al. (2015) developed a multi-agent system using Artificial Bee Colony (ABC) algorithm for the extraction of endmembers. The method solved the problem in high speed computing and distributive environments. Zhang et al. (2017) analyzed the role of swarm intelligence in the extraction of endmembers from hyperspectral images. Swarm intelligence provides a reliable solution. Kusetogullari et al. (2015) proposed Parallel Binary Particle Swarm Optimization for unsupervised change detection and compared the results with other methods. Suresh and Lal (2016) have implemented CS McCulloch for image segmentation. The results were compared with various techniques and were validated against by various measures. Singh et al. (2016) has presented a comparison of Artificial Bee Colony (ABC), Particle Swarm Optimization (PSO) and Cuckoo Search (CS) along with their hybrids for image enhancement. Sood and Menon (2016) proposed a hybrid Bat Algorithm-Cuckoo Search (BA-CS) approach for the discovery of best path for robotic navigation.

Bhandari et al. (2016) performed a comparative study of various wavelet filters for de-noising satellite images using CS, PSO and ABC. Senthilnath et al. (2016) proposed the implementation of Bat Algorithm (BA) in crop classification problem and compared the result with other intelligent algorithms. Gharbia et al. (2016) proposed image fusion method using Particle Swarm Organization (PSO). The method improved the spatial information and preserved spectral resolution. Tebbi and Haddad (2017) have proposed the use of Support Vector Machine (SVM) classifier in satellite image classification and the classification error was considerably reduced. Muangkote et al. (2016) presented an enhanced algorithm for the segmentation of image using Moth-Flame Optimization. The proposed method was more accurate and effective as compared to other traditional methods. Kusetogullari and Yavariabdi (2016) proposed the implementation of Self-Adaptive Hybrid Particle Swarm Optimization-Genetic Algorithm in obtaining change detection for Landsat multi temporal multispectral images. Sarkar et al. (2016) gave a novel unsupervised classification technique using DE and maximum Rényi entropy methods.

Tien Bui et al. (2017) implemented a novel hybrid Neural Fuzzy optimized by Particle Swarm Optimization (PSO-NF) technique for forest fire susceptibility modeling. Bhandari et al. (2017) proposed the implementation of Beta Differential Evolution (BDE) algorithm in image contrast enhancement. The results with respect to SSIM, EKI, MSE, PSNR and FSIM show the superiority over other traditional methods. Sachdeva et al. (2017) proposed a predictive model for flood susceptibility using PSO and SVM. K et al. (2016) reviewed the enactment of PSO and classifier such as Random forest to satellite images for enhancing and obtaining accurate model of Land Cover Classification. Chang et al. (2017) implemented a novel approach for dimensionality reduction. The Impurity Function band prioritization method uses PSO and Gravitational Search Algorithm for reducing the hyperspectral bands. Golovko et al. (2017) has proposed the use of convolutional neural network in low-quality satellite images for detection of solar photovoltaic panels. Google satellite images were used. Azarang and Ghassemian (2017) proposed a novel approach of image fusion for applications in remote sensing using particle swarm optimization for weight injections. WorldView-3 and QuickBird data set are considered for assessment. Kumar et al. (2016) showed implementation of PSO and K-means to cluster satellite images. The approach produced more condensed and augmented clusters than the K-means method alone. Gaba et al. (2017) developed a statistical model, which helps in learning and classifying object in hyperspectral images using combination of GSA and FODPSO. Alizadeh Naeini et al. (2018) considered satellite images of very high spatial resolution and proposed a novel object based feature selection method. Singh et al. (2017) has proposed the use of Moth Flame Optimization for image classification. The detailed comparison is presented in Table 4.

Table 4 Comparative analysis from 2015 to 2018
Full size table

Discussion

It has been observed that different techniques have been applied in different sectors of satellite image processing. In segmentation of satellite images, Particle Swarm Optimization is most widely used followed by Cuckoo Search, Artificial Bee Colony, Differential Evolution, Wind Driven Optimization, Genetic Algorithm and Moth-Flame Optimization. Classification is covered by Particle Swarm Optimization, Unsupervised Artificial Immune Classifier, Ant Colony Optimization, Differential Evolution, Fuzzy-Rough Set, Biogeography Based Optimization, Glowworm Swarm Optimization, Artificial Bee Colony, Neural Network/ Convolutional NN, Bat Algorithm, Support Vector Machine and Moth-Flame Optimization. Feature/ Band Selection use Firefly Algorithm, Differential Evolution, Genetic Algorithm and Particle Swarm Optimization. Extraction of roads or map updation or cross-country path-finding is achieved by Cuckoo Search, Bat Algorithm, Ant Colony Algorithm and Particle Swarm Optimization. Particle Swarm Optimization discovers its use in Image Fusion or Similar Image, Image Restoration, Sub-Pixel Mapping, Rational Function Models and Online Dictionary Learning. For Endmember Extraction, Differential Evolution, Particle Swarm Optimization, Ant Colony Optimization and Artificial Bee Colony are implemented. Genetic Algorithm and Particle Swarm Optimization are implemented for Image Registration and Change Detection. Fire and Flood susceptibility model is implemented by Particle Swarm Optimization and Support Vector machine. Contrast/ Image Enhancement are done by using Differential Evolution, Particle Swarm Optimization, Cuckoo Search and Artificial Bee Colony. Clustering uses Particle Swarm Optimization and Glowworm Swarm Optimization, whereas De-noising is using Artificial Bee Colony, Particle Swarm Optimization and Cuckoo Search. The brief analysis of various artificial intelligent techniques with their application areas in satellite image optimization is depicted in Table 5. The statistical analysis is carried out by pie and column charts as shown in Fig. 1. From Fig. 1a, this is observed that in image segmentation, Particle swarm optimization has the maximum applications (31%) followed by Cuckoo Search (23%), Artificial bee colony algorithm (15%), wind driven optimization, genetic algorithm and moth-flame optimization, differential evolution (7%). From Fig. 1b, PSO has 43% applications with respect to image classification, followed by ant colony optimization, artificial bee colony algorithm and neural networks. Similarly, Fig. 1c, d, e, f, g and h represents the applications of various swarm-based techniques in different areas of image processing. [U13].

Table 5 Image optimization techniques for satellite images
Full size table
Fig. 1
figure 1figure 1

Application areas of Swarm Intelligence Techniques in various fields of Image processing: (a) Segmentation, (b) Classification, (c) feature/band selection, (d) extraction of roads, (e) endmember extraction, (f) contrast/image enhancement, (g) De-noising, fire/flood susceptibility, clustering, registration and change detection, (h) other areas [U14]

Full size image

Conclusion

Remote sensing provides coverage of large areas to collect precise information in various applications such as agricultural fields, location of floods, forest fires, landscape and regional planning etc. The quality of satellite images is weather dependent and size of data base is huge. This makes the image processing task highly time consuming and cumbersome. Thus, there is a need for application of a suitable artificial intelligence technique to improve the image quality with smaller processing time. In this study, various artificial intelligence techniques such as PSO, ACO, ABC, bat algorithm, GA etc. have been analysed for optimization of satellite image data. The detailed analysis revealed that Ant Colony Optimization finds its applications in the field of classification and extraction of endmember from hyperspectral data. Bat Algorithm, Artificial bee colony and Neural Networks work for classification and Cuckoo search algorithm deals with segmentation. This has been observed that Particle Swarm Optimization is the most commonly utilized strategy with respect to image classification, segmentation, feature/band selection, enhancement, image fusion, registration and restoration. The major areas covered by various techniques are land use-land cover mapping, crop classification, forest fire susceptibility and flood assessment. In future, these techniques need to be applied to other significant areas such as natural calamities forecast, suggestions for quick and efficient relief operations and estimation of natural resources. The use of hybrid techniques by combining one or more artificial intelligence for remote sensed image optimization can also be performed. [U15].