Fault-Based Geological Lineaments Extraction Using Remote Sensing and GIS—A Review

: Geological lineaments are the earth’s linear features indicating signiﬁcant tectonic units in the crust associated with the formation of minerals, active faults, groundwater controls, earthquakes, and geomorphology. This study aims to provide a systematic review of the state-of-the-art remote sensing techniques and data sets employed for geological lineament analysis. The critical challenges of this approach and the diverse data veriﬁcation and validation techniques will be pre-sented. Thus, this review spanned academic articles published since 1975, including expert reports and theses. Landsat series, Advanced Spaceborne Thermal Emission and Reﬂection Radiometer (ASTER), Sentinel 2 are the prevalent optical remote sensing data widely used for lineament detection. Moreover, Shuttle Radar Topography Mission (SRTM) derived Digital Elevation Model (DEM), Synthetic-aperture radar (SAR), Interferometric synthetic aperture radar (InSAR), and Sentinel 1 are the typical radar remotely sensed data which are widely used for the detection of geological lineaments. The geological lineaments acquired via GIS techniques are not consistent even though a variety of manual, semi-automated, and automated techniques are applied. Therefore, a single method may not provide an accurate lineament distribution and may include artifacts requiring integration of multiple algorithms, e.g., manual and automated algorithms.


Introduction
In the early 20th century the word "lineament" was used to describe earth surface features such as (i) crests of ridges or boundaries of elevated areas, (ii) the drainage lines, (iii) coastlines, and (iv) boundary lines of formations of petrographic rock types, or lines of outcrops [1][2][3][4]. Later this terminology was expanded to cover additional features like valleys and visible lines of fracture or fault breccia zones. Other definitions of lineament are reported by [5]. Lineament is defined as any mappable, simple, or composite linear feature of the earth surface in which the parts are aligned in rectilinear or slightly curvilinear coherent structures characterized by distinct patterns from adjacent features [5,6]. Genetically, three types of lineaments are separated; (i) geological lineaments, (ii) geomorphological or topographic lineaments, and (iii) Pseudo, manmade, or nongeological lineaments [2,7,8] Earth surface linear features (rectilinear or curvilinear) caused by tectonic activities are faults, fractures, joints, or lithological boundaries, and are termed geological lineaments. Topographic lineaments are caused by geomorphological processes, such as drainage systems and ridges. Pseudo or human-made features are roads, railroads, crop field boundaries, or any variations in land use patterns. In satellite images, the geological lineaments feature significantly brighter or darker linear features than the background pixel 1 Blue 0. 45

Remote Sensing Techniques
Considering the literature review since 1975, there is not a single technique of geological lineament detection in remote sensing, however, as a general concept, the geological lineaments are extracted using the three following techniques: • Manual Lineament Extraction • Semi-Automated Lineament Extraction • Automated Lineament Extraction

Manual Lineament Extraction
Manual lineament extraction is performed by visual interpretation of human operators and is well suited for spatial assessment; the extraction is conducted using an image enhancement process e.g., directional filtering, band ratio, transformation, visual interpretation, and manual digitization of the lineaments [36,52,53]. Manual lineament extraction is applied when the main objective is to demarcate the geological features [54]. The output comes from manual lineament extraction and depends on analyst skills and area of interest. The earliest lineament interpretation was performed using stereoscopic aerial photos, where the lineaments were delineated on the transparent overlays and then transferred to a map [31,55,56]. Digital Terrain Model (DTM) or high-resolution satellite images are preferred by many researchers for the manual extraction of lineaments [54,57]. Lineaments are then extracted by the visual interpretation of an image after it is enhanced. Advances in hardware and technology lead to improved visualization via photo interpretation. Acquired lineaments are reconstructed on the image as hard copies or on the screen digitally [6]. The critical aspects of visual interpretation are generally tonal contrast and textural pattern in which the geomorphological characteristics e.g., drainages pattern and density, rock resistance, landforms, and development of bedding, and superficial cover such as vegetation and cultivation can be reflected in the images [31,58].
In satellite imagery, lineaments usually appear as straight lines or edges that differentiate by the tonal gradients of the surface material [59]. The user's field experience is essential for the accurate classification of lineaments and to complete the broken segments into a longer lineament [59,60]. Furthermore, the quality of images is also considered a key factor for better identifying lineaments in manual approach [61]. Some general features are described by Wang & Howarth [23] to help determine lineaments, and they are topographic features such as straight valley, continuous scarps, straight rock boundaries, a systematic offset of rivers, sudden tonal variations, and alignment of vegetation. A continuous straight valley is considered by Koike [62] as the most powerful feature in image processing for lineament detection as satellite images have no direct information on the topography of the area. The early geological lineaments detection method in remote sensing using a manual approach goes back to the Landsat launch in 1972. Landsat 1 was launched on 23 July 1972; it is known as the first Earth-observing satellite. It consisted of four multispectral bands and was operating until 1978 [63]. The most relevant studies between 1975-2021 using manual lineament extraction are listed in (Table A1 in Appendix A).
This study describes the standard enhancement methods of satellite images that significantly contribute to manual lineament extraction.

Enhancement Techniques
Basically, remote sensing sensors record reflected and emitted radiant flux from the earth's surface materials that one material would reflect-a large amount of energy in a particular wavelength. In contrast, another material would reflect less energy in the same wavelength. This mechanism would result in the difference between two types of materials being recorded by the remote sensing sensors [64]. Sometimes different kinds of materials reflect a similar amount of energy which causes low contrast. Therefore, the level of contrast is determined by the variation in the amount of reflected energy. Image enhancement is required during the image processing, as it not only highlights the weak edges or other linear features but also can improve accuracy [65]. Image enhancement is considered one of the main stages of image processing tools for the lineament extraction [66][67][68].
Furthermore, the difference is also determined based on the radiometric range of the remote sensing systems. For example, a common remote sensing system is designed to record a wide range of brightness values (e.g., 0 to 255) for 8 bits, but due to saturation caused by the radiometric sensitivity of a detector, it cannot record the full range of intensities of reflected or emitted energy from the scene, resulting in relatively low contrast imagery [6,64].
In the visual interpretation and manual extraction of lineaments, the imagery is enhanced in the radiometric range. Various digital methods are available for contrast enhancement that generally are classified as linear and non-linear digital contrast enhancement methods. Contrast stretching, minimum-maximum contrast stretch, and piecewise contrast stretch are the common types of linear contrast enhancement, while histogram normalization is the standard type of non-linear enhancement [64].
Several works have been done in manual lineament extraction using contrast enhancement, particularly histogram equalization and stretching for a single band [69,70]. The other relevant works are summarized in (Table A1).

Spatial Convolution Filtering
One of the characteristics of remotely sensed data is spatial frequency; it is defined as the number of brightness value changes per unit distance for any part of an image [64,71]. Two levels of frequency in images are distinct: low-frequency and high-frequency. Suppose there are fewer changes in brightness value over a given area in an image. In that case, this is referred to as a low frequency area, whereas, if there is a dramatic change over short distances, this is considered a high-frequency area [64,72]. In other words, high spatial frequency in an image is associated with frequent changes of brightness with the position. The most common examples of high-frequency data are edges, lines, and some types of noise. Conversely, gradual changes of brightness value in an image represent low frequency [20,73].
Spatial frequency in remote sensing data can be intensified or reduced using two different techniques; spatial convolution filtering, which is based on convolution masks, and Fourier analysis, which automatically separates an image into its spatial frequency components [64]. The review of many studies shows that spatial convolution filtering is commonly used for lineament extraction; therefore, Fourier analysis is out of the scope of this review.
Based on [64,74], spatial convolution filtering is considered as a linear filter for which the brightness value at a certain location in the output image is a function of the weighted average of brightness values located in a particular spatial pattern around the specific location in the input image. This process of the weighted neighboring pixel value is called two-dimensional convolution filtering. This procedure is utilized for the spatial frequency of an image, for instance, high-frequency linear filtering may sharpen the edges in an image, while low-frequency linear spatial filtering is used to reduce noise within an image. Considering the literature [29,60,64,[75][76][77][78][79], four types of spatial convolution filters are common to be used for manual or automated lineament detection: (a) Low pass filter, (b) High pass filter, (c) Linear edge detection, and (d) Non-linear edge detection.

Low-Pass Filtering
The types of image enhancement that reduce an image's high spatial frequency are called low frequency or low pass filters [64,80,81]. These filters are useful for removing the noise in the images [64,82,83]. The simple low pass filter assesses a specific input pixel brightness value and the pixels surrounding the input pixel and outputs a new brightness value that is the mean of this convolution [64,84]. The size of the neighborhood convolution mask or kernel is usually 3 × 3, 5 × 5, 7 × 7, or 9 × 9. A typical example is shown in (Figure 1). Jensen [64] states that the coefficients in a low-frequency convolution mask are usually set equal to 1. The calculation of the low pass filter is done using the equation below. average of brightness values located in a particular spatial pattern around the specific location in the input image. This process of the weighted neighboring pixel value is called two-dimensional convolution filtering. This procedure is utilized for the spatial frequency of an image, for instance, high-frequency linear filtering may sharpen the edges in an image, while low-frequency linear spatial filtering is used to reduce noise within an image.

Low-Pass Filtering
The types of image enhancement that reduce an image's high spatial frequency are called low frequency or low pass filters [64,80,81]. These filters are useful for removing the noise in the images [64,82,83]. The simple low pass filter assesses a specific input pixel brightness value and the pixels surrounding the input pixel and outputs a new brightness value that is the mean of this convolution [64,84]. The size of the neighborhood convolution mask or kernel is usually 3 × 3, 5 × 5, 7 × 7, or 9 × 9. A typical example is shown in (Figure 1). Jensen [64] states that the coefficients in a low-frequency convolution mask are usually set equal to 1. The calculation of the low pass filter is done using the equation below.
Convolution mask template = Low frequency filter = where Ci is the coefficient, Vi is the individual pixel value, and n is the number of pixel covered by the convolution mask. Figure 1. The filtering image mechanism shows the row and column-wise moving a mask or window (modified from [60]). Figure 1. The filtering image mechanism shows the row and column-wise moving a mask or window (modified from [60]).
The filtered image in (Figure 2) is calculated based on the above equations: The filtered image in (Figure 2) is calculated based on the above equations: The new values in the filtered image: It is notable that low pass filters are used less for the manual extraction of lineaments because the output image is only reduced in terms of noise, and the intensity of blurring is high which causes confusion in lineament extraction. Figure 3 illustrates a visual example of a low pass filtering operation on an original image from a residential area in Germany taken by an unmanned aerial vehicle (UAV) derived from [64]. Figure 3a is a normal color image, while Figure 3b is the low pass filtered image that has been applied on the red band of the original image. In the example, the image becomes blurred by suppressing the high-frequency image.

High-Pass Filtering
High-pass filtering is applied to emphasize or increase the high-frequency components of an image whereas it decreases the low-frequency components. Simply, high-pass filtering is calculated by subtracting the low-pass filter (LPF) from twice the value of the original central pixel value [64,84]: The new values in the filtered image: It is notable that low pass filters are used less for the manual extraction of lineaments because the output image is only reduced in terms of noise, and the intensity of blurring is high which causes confusion in lineament extraction. Figure 3 illustrates a visual example of a low pass filtering operation on an original image from a residential area in Germany taken by an unmanned aerial vehicle (UAV) derived from [64]. Figure 3a is a normal color image, while Figure 3b is the low pass filtered image that has been applied on the red band of the original image. In the example, the image becomes blurred by suppressing the high-frequency image. The filtered image in (Figure 2) is calculated based on the above equations: The new values in the filtered image: It is notable that low pass filters are used less for the manual extraction of lineaments because the output image is only reduced in terms of noise, and the intensity of blurring is high which causes confusion in lineament extraction. Figure 3 illustrates a visual example of a low pass filtering operation on an original image from a residential area in Germany taken by an unmanned aerial vehicle (UAV) derived from [64]. Figure 3a is a normal color image, while Figure 3b is the low pass filtered image that has been applied on the red band of the original image. In the example, the image becomes blurred by suppressing the high-frequency image.

High-Pass Filtering
High-pass filtering is applied to emphasize or increase the high-frequency components of an image whereas it decreases the low-frequency components. Simply, high-pass filtering is calculated by subtracting the low-pass filter (LPF) from twice the value of the original central pixel value [64,84]:

High-Pass Filtering
High-pass filtering is applied to emphasize or increase the high-frequency components of an image whereas it decreases the low-frequency components. Simply, high-pass filtering is calculated by subtracting the low-pass filter (LPF) from twice the value of the original central pixel value [64,84]: where BV is brightness value. Based on [64], due to the high correlation of brightness value with a nine element in high-pass filtering, the filtered image will be characterized by the relatively narrow intensity histogram. Furthermore, the standard kernel used in this filter is 3 × 3 with the coefficients below, which results in a sharpening of the edges.
High-pass filtering is commonly used for edge enhancement which increases the chance of lineament extraction using manual interpretation. Typical examples of highpass filtering using 3 × 3 kernel with different coefficients for the geological lineament extraction are used by [9,29,75]. In these examples, the authors have examined this filtering on Landsat and DEMs over the different areas. Qari [75] examined high pass filtering using a 3 × 3 kernel over the Al-Khabt area, Southern Arabian Shield to manually extract geological lineaments. In this example which is shown in (Figure 4), the author applied the filtering to TM band 5 for edge enhancement to achieve high lineament contrast. Manual extraction of geological lineament from the original image ( Figure 4a) seems difficult, hence the image was changed after the filtering was done (Figure 4b). The image became sharp and the features can be clearly highlighted. For better visualization, the author superimposed the filtered image on the TM band 5 (Figure 4c), consequently, the existing lineaments were achieved using manual digitization ( Figure 4d).

Linear Edge Detection Filtering
Usually, edges in an image are sharp changes in brightness value between two adjacent pixels [85]. One of the most common methods in remote sensing for geological lineament extraction is edge detection. This can be achieved by the edge enhancement techniques, making the shapes and details more apparent and more convenient to analyze

Linear Edge Detection Filtering
Usually, edges in an image are sharp changes in brightness value between two adjacent pixels [85]. One of the most common methods in remote sensing for geological lineament extraction is edge detection. This can be achieved by the edge enhancement techniques, making the shapes and details more apparent and more convenient to analyze [64]. Based on [85], the human eye considers a sharp and distinctive line between adjacent regions of a picture, but the presence of noise and coarse resolution results in a blur and decreases the chance of appearing of edges. Therefore, edge enhancement techniques are applied to overcome this visual distinctiveness. The edges of remotely-sensed data with a gray value and which change sharply are a significant clue for the delineation and extraction of lineaments [86][87][88].
Like the previous filters, linear detection filtering is usually applied using a particular convolution weighted mask or kernel. One of the most effective linear enhancement filtering techniques causing the edges to appear as a plastic shaded-relief format is emboss filters. The edges are obtained using emboss east and emboss northwest directions with the coefficient below [64]: Compass gradient masks are another type of linear edge detection filtering which is used for discrete differentiation directional edge enhancement. The Compass filter uses eight different commonly gradient mask directions (N, E, S, W, NE, SE, SW, NW) [64,89]. The different direction masks with their associated coefficients are shown in (Table 2). Table 2. Directional filtering of the compass with their associated coefficient.
Jensen [64] examined typical examples of Emboss and Compass filters on the red band of the original color image with high spatial resolution of a residential area in Germany ( Figure 5a). Jensen applied Emboss Northwest filtering to achieve the plastic shaded-relief for better visualization of edges ( Figure 5b). Figure 5c illustrates the Compass Northeast filtering of the original image. Geosciences 2021, 11, x FOR PEER REVIEW 12 of 33 Another four 3 × 3 alternative directional filters have been proposed by [73] for edge detection with their associated coefficient, as below: The last linear edge detection filtering is referred to as Laplacian filtering, applied to imagery to perform edge enhancement. This filtering is a second derivative edge enhancement filter without regard to edge direction [64]. The 3 × 3 Laplacian filters are described as below with their coefficients: Linear edge enhancement filtering has been widely used for geological lineaments over the last few decades [76,79,90]. For instance, Farahbaksh [79] tested image transformation and enhancement to extract geological lineaments over the Yinnietharra area, Western Australia using Landsat 8 OLI/TIRS data. Farahbakhsh selected the output of Minimum Noise Fraction (MNF) with the highest eigenvalue to apply the Laplacian and directional filtering ( Figure 6). As a result, almost all linear features, including geological lineaments and streams, can readily be extracted from the filtered images. The findings of [79] show that directional filtering has more effective results in the different striking of Another four 3 × 3 alternative directional filters have been proposed by [73] for edge detection with their associated coefficient, as below: The last linear edge detection filtering is referred to as Laplacian filtering, applied to imagery to perform edge enhancement. This filtering is a second derivative edge enhancement filter without regard to edge direction [64]. The 3 × 3 Laplacian filters are described as below with their coefficients: Linear edge enhancement filtering has been widely used for geological lineaments over the last few decades [76,79,90]. For instance, Farahbaksh [79] tested image transformation and enhancement to extract geological lineaments over the Yinnietharra area, Western Australia using Landsat 8 OLI/TIRS data. Farahbakhsh selected the output of Minimum Noise Fraction (MNF) with the highest eigenvalue to apply the Laplacian and directional filtering ( Figure 6). As a result, almost all linear features, including geological lineaments and streams, can readily be extracted from the filtered images. The findings of [79] show that directional filtering has more effective results in the different striking of geological lineaments (Figure 6c-f

Nonlinear Edge Detection Filtering
The non-linear edge enhancement filtering is based on the non-linear combination of pixels. The most common types of this filtering are the Sobel filter, Prewitt filter, and Robert filter [64,73,91,92].
Sobel edge detection filter is followed by the 3 × 3 window numbering and is calculated using the following equation: And V is the individual pixel value. Sobel kernels are calculated in four principal directions (N-S, NE-SW, E-W, and NW-SE) and are applied in a single band, particularly band 7 of ETM+ due to the greater wavelength and lesser effect of moisture due to the minimal effect of atmospheric haze [59,60,70]. The associated coefficient in four directions is shown in (Table 3) [93][94][95][96]. Table 3. Four direction widows for Sobel and Prewitt filters with their associated coefficient.
An example of the capability of Sobel filtering for automatic geological lineaments was carried out within the Ikole/Kabba region, southern Nigeria by [94]  The Prewitt filtering for edge detection contains two groups of 3 × 3 windows, including horizontal and vertical; however, the edges and lineaments in most cases have more than two directions. Therefore, to detect more direction edges, a Prewitt filtering window of six or eight is utilized (Table 3) [97,98]. The Prewitt filtering is more clarified in the study conducted by [98]. Boutrika et al. [98] applied Prewitt filtering on Landsat 7 ETM+ to extract geological lineaments associated with gold mineralization over the Central Hoggar, South Algeria. In this work, Boutrika et al. performed Principal Component Analysis (PCA) on a multispectral band, and then applied Prewitt filtering using a 3 × 3 convolution kernel on PC1 in four directions (N-S, NE-SE, E-W, and NW-SE) (Figure 8). The structures can clearly be distinguished in (Figure 8b), obtained from the overlaying four direction filtered images. The authors highlighted the geological lineaments corresponding to the deep faults on the filtered image (Figure 8c), in which the extracted lineaments are illustrated (Figure 8d).  The Prewitt filtering for edge detection contains two groups of 3 × 3 windows, including horizontal and vertical; however, the edges and lineaments in most cases have more than two directions. Therefore, to detect more direction edges, a Prewitt filtering window of six or eight is utilized (Table 3) [97,98]. The Prewitt filtering is more clarified in the study conducted by [98]. Boutrika et al. [98] applied Prewitt filtering on Landsat 7 ETM+ to extract geological lineaments associated with gold mineralization over the Central Hoggar, South Algeria. In this work, Boutrika et al. performed Principal Component Analysis (PCA) on a multispectral band, and then applied Prewitt filtering using a 3 × 3 convolution kernel on PC1 in four directions (N-S, NE-SE, E-W, and NW-SE) (Figure 8).   The Robert edge detection filtering is performed by a simple, quick, two-dimension spatial gradient measurement on an image. Once this filtering operation is done, the high spatial frequency regions corresponding to edges are highlighted in the image. Robert is using 2 × 2 mask convolution to give edges in x-direction and y-direction, respectively [99][100][101].
Robert mask in X direction = −1 0 0 1 Robert mask in Y direction = 0 −1 1 0 A typical example of Robert filtering applied on the red band of high spatial resolution color image of a residential area, Germany was carried out by [64] (Figure 9a,b). [99][100][101].
A typical example of Robert filtering applied on the red band of high spatial resolution color image of a residential area, Germany was carried out by [64] (Figure 9a,b). The directional and nondirectional filtering is applied for lineament extraction at various scales and disciplines. The results achieved by them are diverse. In terms of geological lineament extraction, directional filtering has more effective results because faults and fractures are usually distributed at different strikes. Therefore, filtering in a particular direction may not give a suitable product. Several studies are concerned about the directional filters for geological lineaments, e.g., [76] tested directional (Sobel) and nondirectional (Laplacian) filters on the particular bands of Landsat ETM and claims that directional filters in NE-SW and NW-SE gave the best results because the prevailing tectonic trends were at the NE-SW and the NW-SE directions considering the tectonic regime of the study area.
Pour & Hashim [35] applied 7 × 7 user-defined directional filtering in the central gold belt, Peninsular Malaysia on PALSAR data to map the lineaments with their specific trends. Four lineaments in total, trending N-S, NE-SW, NNW-SSE, and ESE-WNW, were The directional and nondirectional filtering is applied for lineament extraction at various scales and disciplines. The results achieved by them are diverse. In terms of geological lineament extraction, directional filtering has more effective results because faults and fractures are usually distributed at different strikes. Therefore, filtering in a particular direction may not give a suitable product. Several studies are concerned about the directional filters for geological lineaments, e.g., [76] tested directional (Sobel) and nondirectional (Laplacian) filters on the particular bands of Landsat ETM and claims that directional filters in NE-SW and NW-SE gave the best results because the prevailing tectonic trends were at the NE-SW and the NW-SE directions considering the tectonic regime of the study area.
Pour & Hashim [35] applied 7 × 7 user-defined directional filtering in the central gold belt, Peninsular Malaysia on PALSAR data to map the lineaments with their specific trends. Four lineaments in total, trending N-S, NE-SW, NNW-SSE, and ESE-WNW, were determined, which are compatible with the ground truth studies and the previously mapped faults.
Allou et al. [77] mapped the faults using different directional filters over an area in western Africa. He applied 7 × 7 convolution of Sobel, Prewitt, and Yesou filters in three major directions (N-S to NNE-SSE, N90o to N100o, NW-SE to NNW-SSE); the authors stated that the high density of lineament had been detected using Prewitt filters on panchromatic bands, while the less dense is detectable by Yesou filters. Further works on lineament extraction using directional and nondirectional filters are stated in (Table A1).

Multiband Analysis
The usage of multiband analysis of remote sensing data plays a critical role in the determination of lineaments. Multiband analysis of remote sensing imageries is mostly applied in visual interpretation or manual lineament extraction of geological lineaments. This analysis is impressive when it is integrated with other outputs of enhancement or filtering. The most applied multiband analysis in lineament delineation through photo interpretation are Principal Component Analysis (PCA), False Color Composite (FCC), and Band Ratio (BR).

Principal Component Analysis (PCA)
Principle component analysis is a multivariate statistical technique to reduce the data dimensionality by transforming the selected data into a new principal component axis, generating an uncorrelated image that contains higher contrast than the original bands [102,103]. According to [64], PCA is the spectral enhancement technique of an image by compressing the information of several spectral bands into just two or three transformed principal component images. Singh & Harrison [104] states that new principal component images may be more interpretable than the original data as the multispectral dataset's dimensionality is reduced. Furthermore, the newly generated PCA images have more spectral difference, and the existing objects can be extracted effectively. PCA is more applicable for lithological mapping, and each lithological unit is revealed clearly in PC images; based on the selected minerals, the PC is selected, or sometimes, the composition of several PCs is created. The boundary between two adjacent lithologies may be considered a lineament or sometimes a fault. Therefore, the analyst can easily digitize the edges from PC images.
PCA is mostly used as a supportive method; for instance, [70], in a study, used PCA as multiband analysis to support the results of directional filters for geological lineaments. In this study, band 7 of Landsat TM was used for spatial filtering, and then PCA improved the image. The authors meant to transfer all the spectral information into three bands; after the calculations, the information was accumulated into the three first bands (96.31%). This could help the authors better to interpret the color composites of these three PCs visually. The result of the lineament map by PC composite interpretation shows about a 17% increase with a concentration in a particular direction which causes authors to take the directional filtering into account.
Novak & Soulakellis [28] examined PCA analysis for geomorphic features on Landsat 5 TM over an area in Greece. He visually interpreted the two sets of PCs (PC1, PC2, PC3) and (PC4, PC5, PC6), respectively. 97% of information belongs to the first set. Furthermore, he composed the Red and SWIR-2 with the first two PCs as a false color composite for better visualization. The lineaments detected by this manual method were correlated to the previously mapped faults and were considered new faults, tectonic and lithological contacts. Another study by [90] reveals the PCA method's capability in manual geological lineament extraction using Landsat 5 TM in western Turkey. The author generated the principal component of all six multispectral bands and then found the most loaded PC; finally, he composed three highly loaded PC images to digitize the lineaments. Lastly, the author integrated the results from PCA with the results obtained by edge enhancement filtering. Several similar studies have been conducted, aiming at geological lineament extraction using PCA along with other spatial and spectral enhancement methods, e.g., [79,105,106].

False Color Composite (FCC)
The band combination or color composite is composed of the three various bands in red, green, and blue (RGB) order; if the composed bands are red, green, and blue, the combination is called true color combination as every object is displayed in its natural color. In contrast, the false color combination is the composition of electromagnetic spectrum channels as red, green, and blue (RGB). The resultant image of the object is not visualized in its natural color [107][108][109]. The Optimum Index Factor (OIF) procedure was devised by [110] to rank all the possible combinations of triplets within a given set of spectral bands in order of statistical information contents. In other words, it is the procedure of selection of the best combination of spectral bands for subsequent false color composite. The suitable selection of band composite is also performed using deterministic methods based on the consideration of the targeted objects' spectral characteristics.
Salvi [111] applied false color composite on Landsat TM over Central Italy to interpret the existing geological faults and lineaments. The author utilized the OIF procedure on six VNIR bands to select the best band composite. The OIF ranking result shows that the composition of bands 5, 4, and 1 are suitable to differentiate the objects on the image.
According to [60], false color composite increases the interpretability of the data for manual lineament extraction; among the composites, the author states that the best visual quality is achieved with a false color composite of visible and near IR bands. Therefore, the combination of 432 as red, green, and blue, respectively, for Landsat ETM, was used to identify the linear patterns of vegetations, geologic formation boundary, river channels, and geologic weakness zones.

Band Rationing (BR)
Band rationing is the division of one spectral band by another, based on the spectral characteristics of selected objects. Sometimes similar targeted surficial objects produce different brightness values due to topographic slope and aspect, shadows, or seasonal changes in sunlight illumination angles and density in single or multiple bands, which may confuse the interpreter. Band rationing techniques transform the data, reduce the environmental effects, and provide information that may not be available within a single band [64,112]. BR is calculated as: where BV i,j,r is the output for a pixel at row i and column j; BV i,j,k is the brightness value in band k at the same location and BV i,j,l is the brightness value in band l.
According to [113], several band ratios which are combined in RGB offer greater contrast between the units than the individual band false color images. Furthermore, sometimes band combinations are made of three band ratios for better visualization.
Sarp [60] tested a band ratio of 5/7 in Landsat ETM to remove the effect of shadows, aiming to detect and interpret linear features easily. In most cases, lineament may be the boundary of two adjacent lithologies. Therefore, the author examined three band ratios (5/7, 2/3, 4/5) as an RGB combination to improve visualization in the manual lineament extraction process. The logic behind selecting these band ratios is: 5/7 discriminates the hydroxyl bearing minerals as a good indicator for the water effects along the fracture, 2/3 delineates the contrast between dense and sparse vegetation, and 4/5 highlights the distributed areas in a dark or black tone. Most of the works associated with lineament extraction using multiband analysis during the last few decades are shown (Table A1).

Semi-Automated Lineament Extraction
Semi-automated lineament extraction is the process of lineament detection based on some digital image analysis techniques and visual interpretation. Once the pre-processing and enhancement are performed, the lineaments are extracted using digital image analysis and somehow the manual editing of detected lineaments is also conducted by the user [6,62,70,114]. Concerning manual lineament extraction approaches, this method takes less time and achieves the result faster; however, it may mix the artifact lineament in the resultant map. According to [115]'s statement, semi-automated lineament detection produces more structure than human interpretation, as image processing might be better to enhance the subjected image and linear features. In total, five steps are included in semi-automated lineament extraction: pre-processing, edge detection, edge tracking, edge linking, and vectorization [116]. A bit of enhancement and filtering operations are carried out before the image processing; the enhancement and filtering techniques have been discussed in the previous section. Once the pre-processing is completed, different algorithms, including object-based classification, segmentation, segment tracing algorithm (STA), Hough Transform, and PCI Line, are applied [116][117][118]. Those works which have been done using the semi-automated method are stated in (Table A1).
Mallast et al. [117] have applied semi-automated lineament extraction, combining linear filtering and object-based classification on DEM to delineate the lineaments over the Dead Sea in Israel. The authors operated the 30 × 30 matrix filter and other 5 × 5 s-order filters in four directions to enhance linear features' appearance. Then, they ran the objectbased classification on the generated map using segmentation and vectorization. For a better generation of lineament maps, the authors combined the resultant lineament with the automated lineament extraction Canny algorithms; also, they used ancillary data such as geology and drainages to differentiate the significant lineaments from insignificant.
Lastly, the author concluded that medium resolution DEM is sufficient data for lineament extraction if it is followed by the auxiliary data and highlighted the capability of semiautomated lineament extraction.
A novel study by [119] shows the capability of the semi-automated method using Unmanned Aerial Vehicles (UAVs) for the mapping of geological structures (faults, fractures, and joints). The authors first detected the geological structures over a small area in the 2D environment, then transferred them into a 3D model to calculate the dip and dip direction. 79.8% accuracy of semi-automated methods is reported compared to expert manual digitizing and interpretation methods. Furthermore, the authors estimated that time is reduced from 7 h to 10 min when the geological structures are mapped using the semi-automated method instead of the manual method.
Semi-automated tectonic structure mapping has been applied in a research by [120]. The author developed a semi-automated method by using software "CurvaTool" and user thresholding. Digital Terrain Model (DTM) has been utilized in this study. The author called this methodology semi-automated because two threshold values are asked from the user, including the expected orientation of data sets of lineaments and their variability range. After the edges are extracted, manual editing may be required to achieve a complete and correct set of lineaments. The results achieved by this developed software were compared with literature data for validation, and good correspondence between literature data and CurvaTool was found concerning the geometry and distribution of the lineaments.
A recent work by [118] highlights the objectivity of semi-automated lineament extraction using Object-Based Image Analysis (OBIA) with high-resolution remote sensing and airborne geophysical datasets. The Southwest of England was tested for this analysis. Here two complementary methods (top-down and bottom-up) segmentations were applied using Trimble eCognition software. The results from the two methods are the same. The results were validated by the data obtained during the survey and field works.

Automated Lineament Extraction
Automated lineament extraction is performed using computer-assisted software. The automated processing includes enhancement, filtering, edge detection, and finally, lineament extraction. There are several algorithms developed by scientists which automatically extract the lineaments, e.g., Hough Transform [23], Lineament Extraction and Stripe Statistical Analysis (LESSA) [121], Segment Tracing Algorithm (STA) [62], Canny Algorithm [69], ADALGEO [122], TecLines [123], Lineament Detection and Analysis (LINDA) [124]. These algorithms are designed to operate via a software package and give the final lineament map in vector format. An automated lineament extraction algorithm takes into consideration the noise, threshold, size, and orientation of linear features [125].
The automated lineament extraction methods help the analyst to produce the result in a short time; however, they still will contain some problems. Sometimes the extracted lineaments or linear features do not correspond to geological structures. Therefore, the user or analyst is obliged to evaluate the extracted lineaments or integrate the manual interpretations.
The earliest automated lineament extraction is the application of Hough transform; a technique being used to separate specific shape features in an image. The algorithm is commonly used for the detection of lines, circles, and ellipses [60]. The Hough transform was firstly applied by [23] for the detection of straight lines representing geological lineaments over an area in Sudbury in Canada. The author stated that being relatively unaffected by gaps in lines and by noise are the main advantages of Hough transform algorithm. The procedures involved in this study are finding local maxima, application of an inverse Hough transform, and straight-line profile analysis. Firstly the author pre-processed the image using a 3 × 3 median filter, and an algorithm was applied to trace all the edges in the image; then, Hough transform was applied to the edge image to produce the cumulator arrays where the brighter tones represented higher counts. Once the local maxima was selected, the inverse Hough transform, and straight-line profile analysis were done to obtain the lineaments.
Vassilas et al. [11] also applied the automated lineament extraction using the Hough transform algorithm; however, the algorithm was modified. The authors first performed data clustering using a self-organizing map, then binarized the classification, and finally applied the modified Hough transform in order to identify lineaments around the Vermion area in Greece. Another study by [45] also shows a Hough transform automated lineament extraction algorithm. In this study, the author tested three spatial data JERS-SAR, Landsat TM, and DEM, for automated and manual interpretation. The final lineament map was obtained using the pre-processing, filtered image in binary format, and finally, the Hough transform algorithm application.
Another developed automated lineament extraction algorithm is the Segment Tracing Algorithm (STA) by [62]. This algorithm was commonly used before the year 2010 for the purpose of geological lineament detection. The STA principle is to delineate a line of pixels as vector elements by examining local variance of the gray level in digital images and connecting the line elements along with their expected directions. In a study, Koike et al. [62] described that the advantages of the proposed algorithm over usual filtering methods are its capability to trace only continuous valleys and extract more lineaments parallel to the sun's azimuth and those located in shadow areas.
Masoud & Koike [126] performed an updated application of STA to automatically delineate tectonically significant lineaments with SRTM DEM and geophysical data. In this study, the authors used 30 arc-second DEM, 1-min gravity anomaly girds, and 2-min total field magnetic intensity grids covering Egypt. Using the developed techniques, orientations, and styles of faulting (normal, reverse, and strike-slip types) were detected. Development in programming languages has also promoted the creation of selfprogrammed software regarding the automated geological lineament. One of them is TecLines, a MATLAB-based toolbox developed by [123]. In this toolbox, the authors have integrated several functions e.g., filters in both frequency and spatial domains, tensor voting framework to produce binary edge maps, and the Hough transform algorithm to extract linear image discontinuities; also B-spline as polynomial curve fitting was used to eliminate the artificial line segments, which are out of interest. The results have been compared with Canny and other algorithms, which give considerable results. The TecLines toolbox has two parts; Line segment detection and Extraction & Line Linking and Merging, which have been published in two separate works [123,127].
In addition to other algorithms, wight overlay analysis is also considered a suitable automated geological lineament extraction approach. Abdullah et al. [115] examined an overlay model technique to produce a predictive fault potential map. The authors considered four significant factors: drainage patterns, previously mapped faults, lineaments, and lithological contact layers. The layers have been created using various techniques; among them, lineament maps were achieved automatically via band 5 of Landsat ETM −7 using PCI Geomatica software. Then, the layers were weighted considering their importance in terms of faulting. The final resultant map was classified into five potential zones: very low, low, moderate, high, and very high. Lastly, the fault line was extracted and compared to the data obtained from the field; a high correlation is seen between the data, which shows the technique's capability.
A recent visual basic-based software LiNDA (Lineament Detection and Analysis) was developed by [124]. This software is a graphical user interface that automatically detects and analyzes the linear features from grid data of DEM, gravity, magnetic, and other remote sensing data. This software's most useful ability is to calculate strike, dip, and estimate the fault type with interactive viewing of lineament geometry. The main working scheme of LINDA includes selection of source grid data, segment detection, and grouping based on STA, transformation from segments to lineaments based on B-Splines, identification of fault type, estimation of strike & dip, construction of fault plane, and the final stage of lineament display with fault type, rose diagram of strikes, and Schmidt's net of directional frequency [124].
Canny edge detection algorithm is another powerful automated linear feature detection that was developed by [128]. According to this algorithm, image convolution with a Gaussian kernel is performed to smooth noise and computes the edge strength and direction for each pixel in the smoothed image. Canny algorithm is based on computing the gradient magnitude that is above some threshold, identified as edges. In other words, this algorithm is applied to smooth the image and reduce the noise, and then find the image gradient to highlight regions with high spatial derivatives.
Consequently, the algorithm goes through these regions and overthrows the pixels which are not at the maximum. The gradient is still further reduced by hysteresis. Hysteresis is utilized to go along the remaining pixels that have not been overthrown. Hysteresis uses two thresholds. Suppose the magnitude is below the first threshold. In that case, it is marked to zero (made a non-edge); on the contrary, if the magnitude is above the high threshold, it is made to an image. In contrast, if the magnitude is between the two thresholds, then it is set to zero, representing a path from this pixel to a pixel with a gradient above the threshold [69,79,128].
In recent years, LINE module of Geomatica software has become the common tool for automated geological lineament extraction [79,84,[129][130][131]. The LINE module's logic is the same as the STA algorithm; however, it also worked based on the Canny algorithm in the first stage of operation. The LINE module of PCI Geomatica extracts the line features. It gives the output as vector segments by using the six parameters: Filter Radius (RADI) with recommended values between 3-8, Edge Gradient Threshold (GTHR) with accepted values between 10-70, Curve Length Threshold (LTHR) with the common value of 10 in most cases, Line Fitting Error Threshold (FTHR) with the recommended values between 2-5, Angular Difference Threshold (ATHR) with the suitable angle between 3 to 20 degrees, and Linking Distance Threshold (DTHR) with efficient values between 10 to 45 [34]. These parameters are also operated by three stages (in blue), as depicted in (Figure 10).  [34]. These parameters are also operated by three stages (in blue), as depicted in ( Figure  10).

Discussion and Concluding Remarks
During the last few decades, remote sensing has proven its effectiveness in every branch of geology, particularly in structural geology, tectonics, lithological mapping, and earthquake monitoring. Lineament is a geological feature that includes faults, fractures,

Discussion and Concluding Remarks
During the last few decades, remote sensing has proven its effectiveness in every branch of geology, particularly in structural geology, tectonics, lithological mapping, and earthquake monitoring. Lineament is a geological feature that includes faults, fractures, joints, and lithological boundaries. These structures can predominantly be determined from diverse remote sensing data sources and techniques. As such, the determination of geological lineaments allows the sufficient exploration of mineral deposits associated with hydrothermal alterations and magmatic processes. In addition, the lineaments can act as the influencing factors of groundwater potentiality, natural hazard assessment, including earthquakes and seismicity, and land sliding.
Conventional geological lineament methodology is time-consuming and expensive. On the other hand, conventional approaches are challenging to apply due to the physical/geographical condition of the region of study. Remotely sensed mapping of geological lineaments leads to determine the faults, and their tectonic implication is the substitution of conventional methods. Manual geological extraction is carried out on the computer screen or hard copies. The growth of technology has changed the photo interpretation operation from the early usage of the stereoscope and transparent overlays to a very fast computer screen with HD displays. Manual geological lineament extraction is directly associated with the interpreter's skill and experience and the remote sensing data. As the lineaments, the sharp and distinctive line between adjacent regions of an image, are seen by human eyes, therefore, an image with coarse resolution and noise resulting blur decreasing the chance of appearing edges and/or linear features. To overcome this visual distinctiveness, different edge enhancement and filtering are applied. Ramli et al. [31] stated the manual lineament extraction subjectivity as a controversial result, reproductivity, and positioning in highly vegetated or wide valleys. The authors also described the final lineament map as integrated with the results from multiple observers to minimize the subjectivity.
Automated and semi-automated approaches of the lineament extraction are performed using computer-assisted software, which is considered one of this approach's main objectivities. The results of the automated and semi-automated techniques depend on the remote sensing data and the algorithm to be applied. The remote sensing data are varying regarding the contrast and the resolution. Therefore, various edge enhancement, directional, and nondirectional filtering are applied to data before the extraction algorithm's operation. STA, Hough Transform, LINDA, Curvatool, and TecLines are the standard algorithm which are utilized for the determination of geological lineaments. Extraction of lineaments is an easy task using automated algorithms. However, in most situations, the algorithm generates segmented images containing several spurious lineament pixels that are not corresponding to geological lineaments. Ramli et al. [31] consider this issue as the main problem with automatic lineament extraction caused by the filtering.
Furthermore, the automatic algorithms cannot effectively map the geological lineaments from low-contrast and mountain shadows, resulting in a low density of lineaments [62]. To overcome this problem, Koike et al. [62] proposed the STA non-filtering method, which extracts the lineaments based on the entered thresholds. Currently, the LINE module in PCI Geomatica software is widely used for the lineament extraction, which is based on the STA and Canny algorithms.
The reviewed literature shows that automated and semi-automated techniques have been applied to detect most works' geological lineaments. The author's experience and review of the previous literature results disclose that a single technique may not efficiently determine the exact geological lineaments excluding artifacts. Therefore, it is suggested to generate a significantly tectonic-geological lineament map using the integration of automatic and manual techniques. The generated map should be compared and verified with the previously every possible mapped structure. In case of possibilities, integration of geophysical data and extensive fieldworks is suggested. Geological lineaments are extracted for various purposes such as tectonic faults, hydrogeology, urban studies, earthquakes. In each case, specific validation and accuracy assessment is applied. In targeting the tectonic faults, the generated geological faults are validated using paleoseismic data or previously mapped faults. The generated map is suggested to be validated with the land deformation deduced by the seismic activities to get the confidential result. The seismic activity or land deformation must indicate an active fault. Therefore, geological lineaments' distribution might be observed in an area with high activity of seismicity and land deformation. Assessment of land deformation around the detected lineaments is performed using GPS stations, conventional geological methods, and remote sensing data. Radar data, particularly Interferometric Synthetic Aperture Radar (InSAR), can effectively produce the land deformation assessment maps showing uplift and land subsidence with their associated velocity.
Author Contributions: H.A. contributed the main conceptualization, designed the methodology, investigated literature, analyzed data, wrote the first draft of the manuscript. E.P. supervised the data analysis, draft preparation, review, and edited the manuscript. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.

Acknowledgments:
The authors thank Kerem Pekkan for his constructive comments and suggestions that have developed the manuscript. Also, two anonymous reviewers are thanked for their thoughtful comments and suggestions that helped enrich this work.

Conflicts of Interest:
The authors declare no conflict of interest.