The detection of photovoltaic panels from images is an important field, as it leverages the possibility of forecasting and planning green energy production by assessing the level of energy autonomy for communities. This entry providposes a summary of n automatic approaches proposed in the literature for that can detecting photovoltaic panels from remote sensingconforming to a properly formed significant range of colours extracted according to the given conditions of light exposure in the analysed imagery. These methodologies encompass machine learning, deep learning, spectral informations. The significant range of colours was automatically formed from an annotated dataset of images, and consisted of the most frequent panel colours differing from the colours of surrounding parts. Such colours were then used to detect panels in other images by analysis, andng panel colour analysiss and reckoning the pixel density and comparable levels of light.
1. Introduction
The integration of renewable energy sources, such as solar power harnessed through photovoltaic panels, within the context of a smart grid has contributed to diminished reliance on conventional fossil fuel-based power generation facilities
[1][2]. Photovoltaic (PV) systems are one of the most promising low-carbon energy generation methods
[3]. PV energy production has grown rapidly over the last decade, at a rate of more than 35% annually
[4][5]. At the end of 2022, the world’s cumulative installed PV capacity was 1055.03 GW, compared to 589.43 GW in 2019, almost doubling in three years
[65].
Estimating the total installed PV capacity and power generation can enhance the ability of policymakers and stakeholders to evaluate progress in terms of sustainability, quantify the actual benefits of green energy, and consider potential future installations
[76]. Aerial and satellite images have been analysed to recognise PV panels by means of approaches using machine learning (ML), i.e., convolutional neural networks (CNN)
, and deep learning methods
. [8][9][10][11][12][13][14][15][16][17]However,
ML tools require a
nd random forest large amount of labelled datasets for their training to be effective, and the related effort to build datasets and perform training is
[18][19][20][21]costly and time consuming.
Further approaches have focused on analysing the physical absorption and reflection characteristic of PV panels to identify them from airborne images
[227][238].
AnThe related results sho
ther approach aimw that the identification of shapes and areas of PV panels is not very accurate. In addition, spectral detection is sensitive to the thresholds selected for the spectral bands of the different sensors.
This entry presents a
n approach t
hat identif
yingies PV panels by means of a deterministic algorithm that
carefully and extensively ananalyses the colours of the pixels forming the panels
. [23].
2. Detecting Photovoltaic Panels in Aerial Images
EThe proposed approach first analyses imating the number of ges to reveal potential intermingling between PV panels in a region is a complex task due to the insufficiency (or even lack of) official registers. Many papers have proposed approaches to detect PV systems by colours and surrounding colours, which mostly affects images having a high level of colour saturation. Second, pixel colours in selected PV panels of labelled images are collected and filtered by excluding the colours typical of background or portions surrounding PV panels. Colour tones commonly found in very dark portions (shadows, ground, etc.) as well as very bright portions (roads, roofs, etc.) are excluded from the colours characteristic of PV panels. The resulting colour selection is a set that properly characterises PV panel colours. This set is then used to analysing satellite and aeriale colours in unlabelled images and detect PV panels. Additionally, for the selected colours that appear in analysed images, often using Convolutional Neural Networks (CNN) or Random Forest (RF) cltheir density is evaluated when considering whether a portion of an image contains PV panels.
2. Proposed Approach
The main ssifiers.
Atrategy of the proposed deeapp
neural network model called Faster-RCNN was used to design the identifiroach is to scan the colours of a small number of annotated aerial images and extract from them a set of characterising PV panel (cPV) colours. Then, these selected colours are used as a reference to detect the location
model of PV panels [12]of PV arrays in other (unannotated) aerial images. The approach
conres
isted of two parts: firstembles and extends the methodology presented in[9][10],
a Rwhe
sNet-50 classifier was pretrained, then a CNN was fine-tunedre colour analysis and mobile tile-based detection were employed to automatically determine the green and urban areas in an image and trace the boundaries between them.
The figure below shows the high-level view of the fo
ur
the identification task of rooftop PV main phases of the proposed approach, consisting of: (i) classification, used to identify images featuring Detectable PV (DPV) panels
. Similarly, three convolutional layers and three fully connected layers were used to evaluate the performance of the ident, i.e., those exhibiting colours different enough from the colours of roofs, roads, ground, etc.; (ii) extraction of cPV colours from DPV images; (iii) detection of PVs in unannotated images using a sliding window that marks tiles on the panels’ surface; and (iv) denoising to filter the previous results.
![](/media/common/202312/3_f2_approach_overview_v4-657c47b83a43e.png)
2.1 Classification
The i
mage classification
[8].phase Mco
reover, eight 2D convolutional layers were used to detect mprises three steps: (i) label assessment; (ii) pixel colour extraction and filtering; and (iii) colour analysis and classification.The label assessment step aimed to reveal the accuracy of the pre-existing PV panel
s in residential are annotations in the initial dataset[11], as
; the to achieve the best performance, thirteen architecturedetection results could be misleading if the annotations are not accurate. For this reason, inaccurate labels were
trainidentified and the
most accurcorresponding images were manually removed.
In genera
l, labels in t
e was selectedhe dataset encompass a region that includes PV panels [9].along Owith
er approaches have used InceptionV3, a CNN used for image analysis and object detection, which was fine-tuned for the task of PV panel identification a significant portion of the surrounding area, such as terrain, vegetation, shadows, and so on. As a consequence, dataset labels cannot be reliably employed to guide the extraction of PV colours. For this reason, a strategy was devised to analyse the colours within the area enclosed by labels and automatically exclude the subset of colours that are also commonly encountered outside the annotated [10][11]area. The
s
e approaches were designtrategy operates as follows. Initially, dataset labels are used to
detect PV panels in both residential and non-residential areas; however, due to the lack of PV panel images, data augmentation was performed durpartition each image into two regions: the Label Region (LR) and its complement, referred to as the Background Region (BR). Subsequently, the RGB colour components of the pixels are collected and the colour count for each region is computed, indicating the t
raining process. The framework proposed inotal number of pixels displaying every colour in each of the two regions. This information is aggregated from all images in the dataset and [10]used to build tw
as employed for the deto sets: LR colours and BR colours (see figure below). Then, for each colour within the intersection of
PV panels in Sweden to collect further market statisticthe two sets, a comparison is performed between the corresponding counts. If the ratio between the BR count and the LR count exceeds
[24].a Similathr
ly, an innovative approach was presented to detect rooftop PV panelseshold, that colour is characterised as a background colour and is removed from the LR set. The outcome of this filtering operation is a subset denoted as PV Panel (PVP) colours.
The analysis o
n the three-dimensional (3D) orientation [25]. Thif the PVP colour sets collected from several images revealed significant colour variations
, approach employed the InceptionV3 model to classify images; subsequently, segmentation and geocoding steps were performed to analyse the 3Dpossibly due to differences in illumination conditions and/or PV panel materials. Three classes of PV panels have been identified, based on the properties of their prominent colours: (i) dark panels, which are typically found in images
. ML and deep learning techniques were used for rooftops PV captured in low-light conditions; (ii) grey–white panels, whose shades can result from the panel material or from sunlight reflecting off the panel
s surface; (iii) detect
ion inable panels [13].(DPV), Tth
e k-means approach was appliat includes all images that do not fall into the previous two classes. A strategy was developed to
segment theautomate the classification of the dataset’s images in
order to define the contours ofto such three classes. The strategy is as follows: for each
rooftop, then a support vector machine (SVM) classifierimage, the LR colour set is selected and its intersection with
a CNN was integrated to accurately identify solar PV arrays. A Mask-RCNN was used for segmentthe PVP colour set is determined, resulting in the image-specific PVP colour set (iPVP). The iPVP set comprises all PVP colours found within that particular image along with their respective counts.
To a
ssess t
ion and identification inhe amount of dark pixels, RGB coordinates of the iPVP colours are converted into [14][16][17]. Tthe
s HSL (Hue
approaches applied the object detection technique to reveal PV panels on aerial images, with CNN being fine-tuned to characterise the mask contours used f, Saturation, Lightness) colour space. Next, the lightness component is evaluated; if a colour falls below a threshold, it is designated as dark. If the aggregated count of dark colours surpasses a predetermined fraction of the total iPVP set counts, then the image is classified as dark.
To
r evaluate the a
rrays. A CNN with the VGG16 encoder was presented inmount of pixels displaying white or grey shades, the fact that such colours can be described as having [15];low fisatur
st, image segmentation was peration is leveraged. Consequently, colours in the iPVP set are transformed
to select the suitable portions of solar panels, then the azimuth of the solar arrays was predicted using edge deteinto the HSV (Hue, Saturation, Value) colour space, which in the researchers' analysis enables better classification based on saturation compared to the HSL space. Similar to the previous step, the count of all colours with a saturation component below a threshold is aggregated. If this combined count exceeds a predetermined fraction
and the Hough transform.
Aof the total counts of the iPVP set, the image is dclassif
ferent approach was proposed inied as belonging to the grey–white class. Images not categorised as either [18]dark to
extractr grey–white are automatically classified as DPV image
s.
2.2 Characterising PV (cPV) Colours Extraction
The goal of
eatures such as this phase is to extract colours
, textures, and other patterns from each pixel, then pass them as input to train an RF classifier to identify pixels related to PV arrays. In a similar appro from DPV images and isolate the characterising PV (cPV) colours. Similar to the image classification phase, each DPV image is partitioned into a Label Region (LR) and its complementary Background Region (BR). Then colours are aggregated and their respective counts for all images, which are recorded separately for each
[19]region,
result
he focus was on the identificaing in sets for the LR colours and BR colours. Next, background colours are removed by evaluating the ratio between the BR count and LR count for all colours within the intersection of
water PV systems (WPV); an RF classithe two sets. As a result of these operations, the PVP colour set is obtained.
A f
iurther
with 400 trees was trainfiltering process is applied to
extract pixels related to WPV, then postprocessing was performed to remove noise and rooftopthe PVP colour set with the objective of excluding both dark and low saturation colours. The reason is that, as illustrated earlier, these colours are not representative of PV panels. A
nother pixel-based RF algorithm usegain, this assessment uses thresholds for the lightness component in the HSL colour space and the
L-8 surface reflectance (SR) product to idesaturation component in the HSV colour space.
2.3 PVs Detection
In
iti
fy suitable PV panels [20]. Thally, PV images are analyse
d RF classifier was based oand all pixels with a colour contained in the
Google Earth Engine (GEE) and used to map PV power plants. Similarly, an RF classifier for an Object-Based Image Analysis (OBIA) approach used different combinations of multispectral Sentinel-2 imagery and radar backscatter from Sentinel-1 SAR imagerycPV set are identified and marked. Then the detection of PV panels in the images is further refined by evaluating the colours within the area of a moving square window. Depending on the features of PV installations, multiple passes with progressively larger windows can greatly enhance the effectiveness of PV detection. A different marking colour is used in each refinement run because each successive run assesses the marked colours within an increasingly larger [26]area. I
f a win
dow [21],in RFa classificsubsequent refinement iteration
was combined with a CNN. First, the RF was used to assign a confidence value to each pixel in order to determine the possibility of that pixel belonging tocontains marked pixels but their cumulative count is not sufficient to trigger the threshold, this indicates that a significant number of pixels within the window do not feature cPV colours. Consequently, it is very likely that the previously marked pixels actually correspond to incorrect detections. In this case, the window is not coloured. An example of detection results is shown in the figure below.
2.4 Image Denoising
Pixels a dis
olar PV array; then, a CNN was used to classify 40×40 patches of RGB images to determine whether or not they corresponded to solar PV playing cPV colours might be found in background regions, although less frequently than for actual PV panels. Occasionally, the count of such pixels is sufficient to trigger the moving window threshold, leading to incorrect detections. To address noise, i.e., occurrences of cPV pixels outside of actual panels
. An innovative deep learning technique called EfficientNet-B7 was employed for PV panel, it is observed that panel surfaces typically contain large clusters of marked tiles. Conversely, incorrect detection
s in
[27],the s
hurro
wing better accuracy and efficiency compared to classic CNN approaches. EfficientNet-B7 was used as a backbone encoder to train a U-Net model for segmentingundings often result in isolated coloured pixels or short coloured stripes. Leveraging this observation, the strategy for identifying and eliminating noise involves evaluating the surroundings of a coloured stripe. To achieve this, the same methodology of the moving window is applied, adapting it for the denoising task. The figure below displays example results of image denoising.
3. Conclusions
This res
olar panels.
Spearch prese
cnt
ral characteristics have been investigated to detect PV s an approach that accurately detects the parts of images showing photovoltaic panels
from hyperspectral data. First, the proposed approach automatically [22][23]classifies byimages focusing on the physical absorption and reflection characteristics of PV panels. To handle the material diversity of PV panel types, these studies applied a tailored image spectral library, which together with the hydrocarbon index mitigated the spectral variance caused by theaccording to the brightness and saturation of their colours and selects those images having appropriate brightness and saturation levels. This step is automatic and does not require annotated images. Second, the colours characterising photovoltaic panels are gathered by comparing panel colours to the colours of the background/surroundings. This step uses a small number of annotated images to discover the panel colours. Third, detection
angle.
Anof photovoltaic panels is a
ppruto
ach for matically perform
ied on images using the
detection by extracting and utilising obtained PV-characterising colours
of PV panels has been presented in [33]and their densities without the need for any previous analysis or or annotation.
ItBy defines a deterministic algorithm that carefully and extensively analyses thecomparing colour sets, the proposed approach is able to automatically exclude false positives and false negatives by quantifying colour densities within tiles.
To validate the effec
olourtiveness of the
pixels forming theapproach, experiments were performed using an open-source dataset of aerial images containing photovoltaic panels. The
appresults showed that the proposed approach can
detectaccurately and consistently identify photovoltaic panels
conforming to a properly formed significant range of colours extracted according, achieving high precision and recall values. Notably, unlike many CNN methods, the approach is capable of identifying both the areas and the actual precise surfaces of PV panels, offering a dual identification capability.
Thanks to the
gi
ven conditions of light exposure in thnitial classification phase used in the proposed approach, datasets for different regions can be analysed
successfully, e.g., images
. The significant range of colours is that have been captured at different times of the day with a specific light can be detected and separated from others, then a subset of the whole set can be automatically
formed from an annotated dataset of images, and consists of the most frequent panel colours differing from the identified for PV detection. Moreover, the automatic extraction of distinctive pixel colours characteristic of PV panel models specific to a given region provides a level of tuning, flexibility, and versatility when applying the approach to images with widely varying colours of
spanels and surrounding
parts. Such colours are then used to detect panels in otherareas. Finally, the proposed approach exhibits a fast execution time, as no training phase is required, and works well without a dataset of annotated images
by analysing panel, as the gathered representative colours
and reckoning the pixcan be used for other regions.
A more
l de
nsity antailed compar
able levels of light. The results show that the approach accurately reveals the contours of panels notwithstandingison with the state of the art, along with a description of the experiments conducted and their
shape or the colours of surrounding objects and the environmentresults, can be found in the related paper.