Fig 8 - uploaded by Micael S. Couceiro
Content may be subject to copyright.
Estimated marginal means of the (a) fitness value and (b) CPU processing time. (Dashed line) Level 10. (Dotted line) Level 12. (Solid line) Level 14.
Context in source publication
Context 1
... clearly seen in Fig. 6. Although it is possible to observe significant differences in the global CPU processing time between the FODPSO and the other algorithms, the improvement of the solution is not perceptible. Hence, in the next section, the same analysis will be performed on a hyperspectral image. 2) Second Data Set— Hyperspectral Image: As for the first data set, the CPU processing times in the second test case for each algorithm for 10-, 12-, and 14-level thresholding were calculated as the average value of 40 different runs, and the results are being presented in Table VIII. According to Table VIII, the FODPSO-based method has the least CPU processing time in comparison with other studied methods as was observed for the first data. On the contrary, PSO is the worst method among others in terms of CPU processing time. As can be seen from Table VIII, FODPSO significantly outperforms the PSO-based method, in particular, when the level of segmentation increases. FODPSO improves the result of the PSO-based segmentation method by 119.6 % and 65.1 % in the best and worst cases, respectively. In the same way, the CPU processing time of the FODPSO is considerably less than that for the DPSO and shows an improvement by 7.4 % and 31.5 % for the best and worst cases, respectively. Table IX gives information regarding the average fitness values of 103 data channels in 40 different iterations. As in the case of the first multispectral data set, in the hyperspectral test case, FODPSO finds optimal threshold values which are better than that for the other methods. This shows that FODPSO is able to find optimal thresholds with better fitness values in less CPU processing time compared to the other studied methods. The fitness value of the FODPSO-based method is followed by DPSO, which is more efficient than the conventional PSO. As can be seen from the table, by increasing the level of segmentation, the fitness of FODPSO increases more than the fitness of the other methods. PSO gives almost the same fitness for 10, 12, and 14 levels of segmentation since it is not endowed with any kind of mechanism to improve the convergence of particles when in the vicinities of the optimal solution. Fig. 7 shows 10-level and 14-level FODPSO-based segmented images using a 200 % zoom. As can be seen from the figure, the 14-level-based segmented image [Fig. 7(b)] provides more details than the 10-level segmentation. Similar to the first data set, the assumption of normality for each of the univariate dependent variables was examined using univariate tests of Kolmogorov–Smirnov ( p -value < 0 . 05) [50]–[52]. The assumption about the homogeneity of variance/covariance matrix in each group was examined with the Box’s M test ( M = 1239 . 38 , F (24; 376576 . 64) = 50 . 58 ; p -value = 0 . 001 ). When the MANOVA detected significant statistical differences, we proceeded to the commonly-used ANOVA for each dependent variable followed by the Tukey’s HSD post hoc. The MANOVA analysis revealed that the algorithm type had a very large and significant effect on the multivariate composite ( P illai s T race = 1 . 40 ; F (4; 702) = 405 . 97 ; p -value = 0 . 001 ; P artial Eta Squared η p 2 = 0 . 698 ; P ower = 1 . 0 ). The segmentation level also had a large and significant effect on the multivariate composite ( P illai s T race = 0 . 97 ; F (4; 702) = 165 . 03 ; p -value = 0 . 001 ; η p 2 = 0 . 49 ; P ower = 1 . 0 ). Finally, the interaction between the two independent variables had a very large and significant effect on the multivariate composite ( P illai s T race = 1 . 02 ; F (8; 702) = 91 . 82 ; p -value = 0 . 001 ; η p 2 = 0 . 51 ; P ower = 1 . 0 ). After observing the multivariate significance in the type of algorithm and the segmentation level, a univariate ANOVA for each dependent variable followed by the Tukey’s HSD test was carried out. For the type of algorithm, the dependent variable fitness value presents statistically significant differences ( F (2 , 351) = 1066 . 64 ; p -value = 0 . 001 ; η p 2 = 0 . 86 ; P ower = 1 . 0 ), as well as the dependent variable CPU processing time ( F (2 , 351) = 2309 . 24 ; p -value = 0 . 001 ; η p 2 = 0 . 93 ; P ower = 1 . 0 ). For the segmentation level, the dependent variable fitness value also presents statistically significant differences ( F (2 , 351) = 3907 . 10 ; p -value = 0 . 001 ; η p 2 = 0 . 96 ; P ower = 1 . 0 ), as well as the dependent variable CPU processing time ( F (2 , 351) = 77 . 58 ; p -value = 0 . 001 ; η p 2 = 0 . 66 , P ower = 1 . 0 ). Using the Tukey’s HSD post hoc, one can observe that there are statistically significant differences between experiments using the PSO, DPSO, and FODPSO segmentation algorithms, for both CPU processing time and fitness function. Once again, the FODPSO produces better solutions than both the PSO and the DPSO in terms of fitness value. Furthermore, as expected, the DPSO produces better solutions than the PSO. As shown in Table X (also shown in Fig. 8), based on Tukey’s HSD post hoc test, the fractional-order algorithm is able to once again reach a better fitness solution in less time. Moreover, the differences between the FODPSO and the other algorithms are more evident as the segmentation level increases. This should be highly appreciated as many applications require real-time multisegmentation methods (e.g., autonomous deployment of sensor nodes in a given environment). In summary, it is possible to observe that the FODPSO is faster than the DPSO since fractional calculus is used to control the convergence rate of the algorithm. As described in [49], a swarm behavior can be divided into exploitation and exploration . The exploitation behavior is related with the convergence of the algorithm, allowing a good short-term performance. However, if the exploitation level is too high, then the algorithm may be stuck on local solutions. On the other hand, the exploration behavior is related with the diversification of the algorithm which allows exploring new solutions, thus im- proving the long-term performance. However, if the exploration level is too high, then the algorithm may take too much time to find the global solution. In the DPSO, the tradeoff between exploitation and exploration can only be handled by adjusting the inertia weight w . While a large inertia weight improves exploration activity, the exploitation may be improved using a small inertia weight. Since the FODPSO presents a fractional calculus strategy to control the convergence of particles with memory effect, the coefficient α allows providing a higher level of exploration while ensuring the global solution of the algorithm (cf., [38]). IV. C LASSIFICATION Although the main idea behind this paper is to introduce a thresholding-based segmentation technique, it is of interest to see the effectiveness of the new segmentation method on classification. In this way, this section presents a novel framework to prove the efficiency of the proposed method for classification. The proposed classification method is based on the FODPSO and the SVM classifier. Since we do not have reference samples for the first data set, the classification is only performed on the second data set. Fig. 9 shows the general idea of the proposed classification approach. As can be seen, the data have been first classified with SVM and a Gaussian kernel. The hyperparameters have been selected using five-fold cross validation. Each variable has been scaled between − 1 and 1. To carry out a fair evaluation, the input is classified only once, while the output of this step is used for all different levels. By doing that, the accuracy of the classification for different methods is only dependent on the effect of the segmentation method. In parallel, the input data are transformed using the principal component analysis (PCA), and the first principal component (PC) is kept since most of the variance is provided by that. The output of this step is segmented by the proposed FODPSO method. In the final step, the results of the SVM and the FODPSO are combined by using majority voting (MV). Fig. 10 depicts the general idea of the proposed approach with MV. The output of the segmentation methods is a few number of objects, and each object consists of several pixels with the same label. In other words, pixels in each object share the same characteristics. To perform the MV on the output of the segmentation and classification steps, counting the number of pixels with different class labels in each object is first carried out. Subsequently, all pixels in each object are assigned to the most frequent class label for the object. In the case where two classes have the same (most frequent) proportions in one object, the object is not assigned to any of those classes, and the result of the traditional SVM is considered for each pixel in the object directly. The procedure of the new classification approach is described step by step as follows: 1) The input data are classified by SVM. 2) The input data are transformed by PCA, and the first PC is kept. 3) The output of step 2 is segmented by FODPSO. 4) The results of steps 1 and 2 are combined using MV. Fig. 11 illustrates the classification map of the standard SVM and the proposed classification method with 10-, 12-, and 14-level segmentation by FODPSO. The output of the SVM presents a lot of noisy pixels which decrease the accuracy of the classification. The results of the overall accuracy and kappa coefficient for the SVM and the new method with 10, 12, and 14 levels are shown in Table XI. For a better understanding, the classification accuracy for each class is also included in the table. All three segmentation levels improve the result of the SVM classification. The accuracy increases when the number of levels increases from 10 to 14. The main reason behind that phenomenon is denoted as under segmentation in which several objects are merged into a single one. This problem can be ...
Similar publications
Inspired by biosilicification of glass sponges, we designed a catalytic peptide, which formed silica structures in the imidazole-buffered solution. The peptide was adsorbed selectively onto the surface of yeast cells, and the bioinspired silicification led to the formation of a cytoprotective silica shell on individual yeast cells.
This paper provides new insights on the convergence of a locally connected network of pulse coupled oscillator (PCOs) (i.e., a bioinspired model for communication networks) to synchronous and desynchronous states, and their implication in terms of the decentralized synchronization and scheduling in communication networks. Bioinspired techniques hav...
This paper presents the first implementation results of a novel
Unitronics (Unicellular Electronics) architecture based system
that uses a bio-inspired prokaryotic model. It is a programmable
cellular FPGA-like system inspired by unicellular bacterial
organisms, and transposes self-healing and fault tolerant
properties of nature to electronics syst...
Bioinspired design is becoming an increasingly popular approach which makes use of nature as a source of inspiration to develop engineering, architectural and other designs. Tools and methods are currently being analyzed and tested to determine how designers may best leverage biological systems to generate designs. In this paper we present a study...
The leaves of Zea mays Linn. contain abundant nitrogen source and exhibit an extremely high light-harvesting efficiency due to its hierarchical internal structures. In this study, Z. mays Linn. leaves are used as template to generate Au/N-TiO2 nanocomposites through soakage, calcination and Au-nanoparticle-loading processes. Such nanocomposites rep...
Citations
... Different from other tasks in remote sensing, such as object detection [20,21], classification [22,23], and segmentation [24,25], the RSIC task involves identifying the main objects and understanding the relationships of these objects within specific scenes simultaneously. Inspired by natural image captioning (NIC) [26,27], the existing RSIC approaches follow the general architecture of the encoder-decoder framework. ...
In the field of remote sensing image captioning (RSIC), mainstream methods typically adopt an encoder–decoder framework. Methods based on this framework often use only simple feature fusion strategies, failing to fully mine the fine-grained features of the remote sensing image. Moreover, the lack of context information introduction in the decoder results in less accurate generated sentences. To address these problems, we propose a two-stage feature enhancement model (TSFE) for remote sensing image captioning. In the first stage, we adopt an adaptive feature fusion strategy to acquire multi-scale features. In the second stage, we further mine fine-grained features based on multi-scale features by establishing associations between different regions of the image. In addition, we introduce global features with scene information in the decoder to help generate descriptions. Experimental results on the RSICD, UCM-Captions, and Sydney-Captions datasets demonstrate that the proposed method outperforms existing state-of-the-art approaches.
... Although PSO technique has been used for solving multilevel image thresholding problem in the past, very few works have used DPSO and FODPSO algorithms for multilevel image thresholding [3,4,17]. In addition, these algorithms have not been used widely in histopathology image segmentation. ...
... The values of other parameters of these algorithms such as inertial factor, cognitive weight, social weight, velocity range, no. of swarms, swarm range, population range and fractional coefficient were fixed based on the analysis of these algorithms from previous works [3,17,38,53]. Parameters in PSO are usually determined depending on the specific problems, application experience, and numerous experiment tests such as trial-and-error methods, so it has no versatility [53]. Table 5 gives the segmentation results of PSO, DPSO and FODPSO algorithms for all the images in the colonoscopy data set. ...
Histopathology image segmentation is a challenging task in medical image processing. This work aims to segment lesion regions from colonoscopy histopathology images. Initially, the images are preprocessed and then segmented using the multilevel image thresholding technique. Multilevel thresholding is considered an optimization problem. Particle swarm optimization (PSO) and its variants, darwinian particle swarm optimization (DPSO), and fractional order darwinian particle swarm optimization (FODPSO) are used to solve the optimization problem and they generate the threshold values. The threshold values obtained are used to segment the lesion regions from the images of the colonoscopy tissue data set. Segmented images containing the lesion regions are then postprocessed to remove unnecessary regions. Experimental results reveal that the FODPSO algorithm with Otsu’s discriminant criterion as the objective function achieves the best accuracy, Dice and Jaccard values of 0.89, 0.68 and 0.52, respectively, for the colonoscopy data set. The FODPSO algorithm also outperforms other optimization methods such as artificial bee colony and the firefly algorithms in terms of the accuracy, Dice and Jaccard values.
... Image segmentation focuses on analysing and interpreting the images. Many image segmentation methods have been developed by using PSO algorithms for improving the performance of image segmentation [144,145,147]. ...
... In [145], a multi-level thresholding method has been proposed for image segmentation, where a fractional-order Darwinian PSO algorithm is proposed to improve the accuracy of image segmentation. In [144], PSO has been combined with the hidden Markov random field model to improve the quality of the image segmentation. A PSO-based image segmentation method has been introduced in [147], where PSO is embedded into a region-based image segmentation (also named seeded region growing) method for oriented segmentation. ...
Survey/review study A Survey of Algorithms, Applications and Trends for Particle Swarm Optimization Jingzhong Fang 1, Weibo Liu 1,*, Linwei Chen 2, Stanislao Lauria 1, Alina Miron 1, and Xiaohui Liu 1 1 Department of Computer Science, Brunel University London, Uxbridge, Middlesex, UB8 3PH, United Kingdom 2 The School of Engineering, University of Warwick, Coventry CV4 7AL, United Kingdom * Correspondence: Weibo.Liu2@brunel.ac.uk Received: 18 October 2022 Accepted: 28 November 2022 Published: Abstract: Particle swarm optimization (PSO) is a popular heuristic method, which is capable of effectively dealing with various optimization problems. A detailed overview of the original PSO and some PSO variant algorithms is presented in this paper. An up-to-date review is provided on the development of PSO variants, which include four types i.e., the adjustment of control parameters, the newly-designed updating strategies, the topological structures, and the hybridization with other optimization algorithms. A general overview of some selected applications (e.g., robotics, energy systems, power systems, and data analytics) of the PSO algorithms is also given. In this paper, some possible future research topics of the PSO algorithms are also introduced.
... Ghamisi et al. later used these ideas in combination with random forest for classifying spectral images [73]. It is used in [74] again for spectral images, i.e., for the multilevel segmentation of spectral images later to be processed by support vector machines. Of course, we could also list these tasks in preprocessing, but since particle swarm is a classical optimization technique, we decided to use them in this category, i.e., optimization. ...
Fractional calculus has gained a lot of attention in the last couple of years. Researchers have discovered that processes in various fields follow fractional dynamics rather than ordinary integer-ordered dynamics, meaning that the corresponding differential equations feature non-integer valued derivatives. There are several arguments for why this is the case, one of which is that fractional derivatives inherit spatiotemporal memory and/or the ability to express complex naturally occurring phenomena. Another popular topic nowadays is machine learning, i.e., learning behavior and patterns from historical data. In our ever-changing world with ever-increasing amounts of data, machine learning is a powerful tool for data analysis, problem-solving, modeling, and prediction. It has provided many further insights and discoveries in various scientific disciplines. As these two modern-day topics hold a lot of potential for combined approaches in terms of describing complex dynamics, this article review combines approaches from fractional derivatives and machine learning from the past, puts them into context, and thus provides a list of possible combined approaches and the corresponding techniques. Note, however, that this article does not deal with neural networks, as there is already extensive literature on neural networks and fractional calculus. We sorted past combined approaches from the literature into three categories, i.e., preprocessing, machine learning and fractional dynamics, and optimization. The contributions of fractional derivatives to machine learning are manifold as they provide powerful preprocessing and feature augmentation techniques, can improve physically informed machine learning, and are capable of improving hyperparameter optimization. Thus, this article serves to motivate researchers dealing with data-based problems, to be specific machine learning practitioners, to adopt new tools, and enhance their existing approaches.
... Ghamisi et al. later used these ideas in combination with Random Forest for classifying spectral images [72]. It is used in [73] again for spectral images, i.e., for multilevel segmentation of spectral images later to be processed by support vector machines. Of course, we could also list these tasks in preprocessing, but since particle swarm is a classical optimization technique, we decided to use them in this category, i.e., optimization. ...
Fractional calculus gained a lot of attention in the last couple of years. Researchers discovered that processes in various fields follow rather fractional dynamics than ordinary integer-ordered dynamics, meaning the corresponding differential equations feature non-integer valued derivatives. There are several arguments for why this is the case, one of them being that fractional derivatives’ inherit spatiotemporal memory and/or the ability to express complex naturally occurring phenomena. Another popular topic nowadays is machine learning, i.e., learning behavior and patterns from historical data. In our ever-changing world with ever-increasing amounts of data, machine learning is a powerful tool for data analysis, problem-solving, modeling, and prediction. It further provides many insights and discoveries in various scientific disciplines. As these two modern-day topics provide a lot of potential for combined approaches to describe complex dynamics, this article reviews combined approaches of fractional derivatives and machine learning from the past, puts them into context, and thus provides a list of possible combined approaches and the corresponding techniques. Note, however, that this article does not deal with neural networks, as there already is profound literature on neural networks and fractional calculus. We sorted past combined approaches from the literature into three categories, i.e., preprocessing, machine learning & fractional dynamics, and optimization. The contributions of fractional derivatives to machine learning are manifold as they provide powerful preprocessing and feature augmentation techniques, can improve physically informed machine learning, and are capable of improving hyperparameter optimization. Thus, this article serves to motivate researchers dealing with data-based problems, to be specific machine learning practitioners, to adopt new tools and enhance their existing approaches.
... where the low and high thresholds, Th 1 and Th 2 were determined by the fractional-order Darwinian particle swarm optimization (FODPSO) thresholding [36] due to its efficiency and fast process. The change map, CM, was further up-sampled to the original size of input images using the nearest-neighbor interpolation and indexed by CM U . ...
There was an error in the original publication [...]
... Reducing the dimension of hyperspectral data is very important. The fractional-order differential is another conducive method to mining details [35][36][37]. Wang et al. found that the optimal inversion model for soil organic matter in China's Aibi Lake was gray relational analysis-back propagation neural network (GRA-BPNN) with 1.2 order data source [38]. ...
Most of the world’s saline soils are found in arid or semiarid areas, where salinization is becoming serious. Ground laboratory hyperspectral data (analytical spectral devices, ASD) as well as spaceborne hyperspectral data, including Gaofen-5 (GF-5) and Huanjing-1A (HJ-1A), provide convenient salinity monitoring. However, the difference among ASD, GF-5, and HJ-1A spectra in salinity monitoring remains unclear. So, we used ASD, GF-5, and HJ-1A spectra as data sources in Gaotai County of Hexi Corridor, which has been affected by salinization. For a more comprehensive comparison of the three spectra datum, four kinds of band screening methods, including Pearson correlation coefficient (PCC), principal component analysis (PCA), successive projections algorithm (SPA), and random forest (RF) were used to reduce the dimension of hyperspectral data. Particle swarm optimization (PSO) was used to improve the random initialization of weights and thresholds of the back propagation neural network (BPNN) model. The results showed that root mean square error (RMSE) and determination of the coefficients (R2) of models based on ASD and HJ-1A spectra were basically similar. ASD spectra (RMSE = 4 mS·cm−1, R2 = 0.82) and HJ-1A (RMSE = 2.98 mS·cm−1, R2 = 0.93) performed better than GF-5 spectra (RMSE = 6.45 mS·cm−1, R2 = 0.67) in some cases. The good modelling result of HJ-1A and GF-5 data confirmed that spaceborne hyperspectral imagery has great potential in salinity mapping. Then, we used HJ-1A and GF-5 hyperspectral imagery to map soil salinity. The results of GF-5 and HJ-1A showed that extremely and highly saline soil mainly occurred in grassland and the southern part of arable land in Gaotai County. Other lands mainly featured non-saline and slightly saline soil. This can provide a reference for salinity monitoring research.
... It solves the local optimum problem of previous evolutionary algorithms as well as the Otsu method's fixed threshold problem. The authors [32] extended FODPSO to optimise multilevel thresholding on HSI, inspired by the aforementioned approach. When the object of interest did not appear as a peak in the histogram, it suffered from inhomogeneity and was unable to segment well. ...
Hyperspectral Images, which are high-dimensional in nature and capture bands over hundreds of wavelengths of the electromagnetic spectrum. These images have piqued researchers’ curiosity in the last two decades. The purpose of this paper is to investigate how researchers segmented and classified Hyperspectral Images with unbalanced data and few labelled training examples. For the sake of comprehension, the background of Hyperspectral Images and segmentation techniques is briefly discussed at first. The study is organised around different Hyperspectral Image processing techniques such as thresholding, clustering, watershed, deep learning, and other methods. The recent trends and developments in HSI segmentation have been reviewed and compiled using benchmark datasets such as Indian Pines, Salinas Valley, Pavia University, and others. Finally, it is intended that the readers will gain a thorough understanding of existing segmentation techniques, their performance, and fresh research areas for HSI that need to be studied or explored.
... To overcome the concerned complexity issues of multilevel thresholding and achieve optimum multilevel threshold values, over the years, many metaheuristic optimization algorithms have been successfully used in multilevel thresholding segmentation problems for determining the optimum threshold values (Peng et al. 2015), such as differential evolution (DE) algorithm (Cuevas et al. 2010;Sarkar et al. 2011;Sarkar and Das 2013), genetic algorithm (GA) (Hammouche et al. 2008;Maulik 2009), particle swarm optimization (PSO) (Feng et al. 2005;Maitra and Chatterjee 2008;Gao et al. 2010;Ghamisi et al. 2014), artificial bee colony (ABC) algorithm (Horng 2011;Ma et al. 2011;Zhang and Wu 2011;Cuevas et al. 2012), glowworm swarm optimization (GSO) algorithm (Horng 2013;Qifang et al. 2014), cuckoo search (CS) algorithm (Brajevic et al. 2012;Agrawal et al. 2013) and Bat algorithm (BA) (Zhiwei et al. 2015;Zhou et al. 2015). ...
Color image multilevel thresholding segmentation based on metaheuristic optimization algorithm and histogram has been widely used in many fields, but histogram neglects the relationships between the neighborhood pixels and any correlation among gray levels of the image when selecting the optimum threshold values. Therefore, in this paper, a novel and efficient color image multilevel thresholding method is proposed which uses an energy function to generate the energy curve of an image by considering spatial contextual information of the image. The presented color image segmentation technique based on energy curve takes the between class variance, Tsallis entropy and Kapur’s entropy as objective functions. In order to further enhance the segmentation performance, an improved firefly algorithm (IFA) is presented in this paper. The IFA algorithm based on energy function is used for color image multilevel thresholding problem and compared with modified firefly algorithm (MFA), cuckoo search (CS), grasshopper optimization algorithm (GOA), Harris hawks optimization (HHO), emperor penguin optimization (EPO) algorithms. The experimental results are presented in terms of optimal threshold value, optimal objective function values, peak signal to noise (PSNR), structural similarity index (SSIM), standard deviation of the objective values and statistical results. The experimental results show that the presented method outperforms the other algorithms and Kapur’s entropy based on energy curve is better than between class variance and Tsallis entropy for color image multilevel thresholding segmentation.
... Then pan-sharpening is also an image fusion technique, to obtain an image that has high-resolution in both spatial and spectral domain. Such an image has a wide range of applications, such as image classification [1,2], change detection [3], [4] and segmentation [5]. ...
The fusion of multispectral and panchromatic images is always dubbed pansharpening. Most of the available deep learning-based pan-sharpening methods sharpen the multispectral images through a one-step scheme, which strongly depends on the reconstruction ability of the network. However, remote sensing images always have large variations, as a result, these one-step methods are vulnerable to the error accumulation and thus incapable of preserving spatial details as well as the spectral information. In this paper, we propose a novel two-step model for pan-sharpening that sharpens the MS image through the progressive compensation of the spatial and spectral information. Firstly, a deep multiscale guided generative adversarial network is used to preliminarily enhance the spatial resolution of the MS image. Starting from the pre-sharpened MS image in the coarse domain, our approach then progressively refines the spatial and spectral residuals over a couple of generative adversarial networks (GANs) that have reverse architectures. The whole model is composed of triple GANs, and based on the specific architecture, a joint compensation loss function is designed to enable the triple GANs to be trained simultaneously. Moreover, the spatial-spectral residual compensation structure proposed in this paper can be extended to other pan-sharpening methods to further enhance their fusion results. Extensive experiments are performed on different datasets and the results demonstrate the effectiveness and efficiency of our proposed method.
