BiLSTM CLASSIFIER MODEL FOR LAND COVER CHANGE DETECTION AND CLASSIFICATION

Objective: This research proposes a change detection in the satellite images and land cover analysis using a Bidirectional Long Short-Term Memory (BiLSTM) model and vegetation index-based feature maps for future map generation. Change detection complements land cover analysis by identifying and quantifying alterations in land cover over time. Theoretical framework: Land cover analysis and land cover change detection plays crucial role in understanding and monitoring the Earth's surface.Despite its importance, land cover analysis and change detection face several challenges. One major challenge is obtaining accurate and up-to-date information on land cover and change detection across large areas can be complex and costly. Inconsistent data sources and limited access to historical records can hinder the accuracy and reliability of change detection analyses.These challenges requires a combination of technological advancements, improved data availability, refined algorithms, robust validation approaches, and interdisciplinary collaborations. Method: BiLSTM method is used for implementation which is powerful for land cover classification, as it allowing the integration of spatial and temporal information and capturing complex patterns in satellite imagery data. BiLSTM, tend to be more complex but they often offer higher accuracy and the ability to learn intricate patterns and representations. Results and conclusion: The six vegetation index-based feature maps are considered.Therefore, the resulting accuracy is also determined using the Flamingo-Hyena optimization (FHO), and the experimental outcomes disclosed that the proposed model is superior to the existing model in terms of accuracy with 0.95%, MSE with 0.05%, precision with 0.94%, Recall with 0.94%, and F1 measure with 0.94% respectively. vegetation index-based feature maps are essential for land cover analysis and change detection as they provide valuable information on vegetation dynamics, ecological processes, land management, and climate change impacts. Research implications: This process land cover analysis and change detection helps to detect deforestation, urban expansion, agricultural expansion, and other land cover changes. By harnessing these techniques, policy makers can address emerging issues such as deforestation, loss of biodiversity, and encroachment on natural habitats.


INTRODUCTION
Land cover analysis and change detection play a crucial role in understanding the dynamic nature of our planet's surface and its impact on various environmental and socio-economic processes.
Land cover analysis often involves the use of remote sensing data, such as satellite imagery, to classify and monitor different land cover types.
Satellite imagery poses several challenges for land cover analysis and change detection.
These challenges include varying spatial and temporal resolutions/variability, the presence of cloud cover and atmospheric conditions, the need for preprocessing and calibration, managing large data volumes, and ensuring accurate interpretation and validation.Overcoming these challenges requires collaboration, technological advancements, and improved data availability to enhance the accuracy and usability of satellite imagery for land cover analysis and change detection applications.
Vegetation index-based feature maps also plays a crucial role in land cover analysis and change detection due to their importance in assessing vegetation dynamics and ecological processes.4 The basic study of the ecological environment of the Earth often involves the classification of a green environment, with an emphasis on understanding land cover change.[3] There are several methods used for land cover classification, including both traditional and machine learning-based approaches.Here are some commonly used methods for land cover classification with their advantages and limitations The main purpose of the this research is to detect the changes in satellite images between successive years, land cover, as well as landfill using two satellite images of the varied period.
BiLSTM can capture dependencies in both directions, enabling the model to learn from the past and future observations, which enhances the classification accuracy by incorporating the temporal dynamics of land cover changes.
BiLSTM can work better for temporal dependencies and sequential patterns, making it suitable for land cover classification tasks that require considering the temporal dynamics of land cover changes.BiLSTM's ability to handle variable-length sequences and learn robust representations from satellite imagery data is advantageous.
The information required for the change detection of land cover is gathered, and then the preprocessing is performed to create the data suitable for the change detection with a model based on pre-trained deep learning as well as vegetation index-based feature maps will be implemented and the final implementation for the change detection includes Flamingo-Hyena optimization (FHO) and BiLSTM classifier model (FHO-BiLSTM).The flamingo-hyena optimization is developed as new hybrid optimization designed mathematically and theoretically with the features of foraging, migratory, and hunting capability of Phoenicopteridae [18] and Hyaenidae [19].The BiLSTM classifier model is proposed for detecting the changes in satellite images, and the generated feature maps are subjected to the classifier, which employs the Adam optimizer to improve the tuning performance of hyperparameters.

CHALLENGES
Land cover refers to the physical and biological cover of the Earth's surface.Analyzing land cover and detecting its changes provide valuable insights into environmental processes.However, these tasks come with their fair share of challenges.Some of the key challenges include data quality, classification accuracy, scale and resolution, change detection algorithms, uncertainty quantification, interpretability and explainability, multi-temporal analysis, transferability and generalization, ancillary data integration, data heterogeneity and interoperability, and computational efficiency.
From traditional approaches to advanced remote sensing techniques, a range of methods are developed and employed to assess and monitor these changes for land cover change detection and analysis.These methods encompass diverse tools and technologies, including satellite imagery, aerial photography, geographic information systems (GIS), and machine learning algorithms.Each method offers unique advantages and challenges, making it essential to consider the specific research objectives, data availability, and spatial and temporal scales when selecting the most appropriate approach.highlighting their strengths and applications in monitoring.utilized.BiLSTM (Bidirectional Long Short-Term Memory) is a type of recurrent neural network (RNN) that can be applied to address these challenges in land cover analysis and change detection.BiLSTM networks have the ability to model sequential data and capture longterm dependencies, making them suitable for analyzing time-series data such as satellite imagery.This method can capture complex spatial and temporal relationships, handle largescale datasets, and automatically learn relevant features from the data.They offer the potential to address challenges related to classification accuracy, change detection, scale, and data integration.However, these techniques also require adequate training data, computational resources, and careful model design and validation to ensure robust and accurate results.

Model
The main intention of this work is to detect the changes in the satellite images between successive years as well as to detect the changes in the land cover as well as landfill.Here, two satellite images of varied periods are utilized, and the collected satellite images are required to be preprocessed due to the presence of noises from haze, clouds, etc.When these unwanted artifacts are removed from the images; then they are subjected to feature map generation.In the feature map generation, a pre-trained deep learning-based model, as well as six vegetation index-based feature maps, will be implemented in which six vegetation indexes are considered for the establishment of the features are the Normalized Difference Vegetation Index(NDVI), Soil-Adjusted Vegetation Index(SAVI), Weighted Vegetation Index (WVI), Global Environmental Monitoring Index(GEMI), Modified Soil-Adjusted Vegetation Index (MSAVI), and Modified Simple Ratio(MSR).After generating the feature map, the satellite images will be subjected to the BiLSTM for detecting the changes as the model considers long-term province of bi-directional among time steps of time series or data patterns in satellite images.In this research, the novelty is concentrated on obtaining the classification accuracy enhancement for which the classifier training will be strengthened.A new meta-heuristics model instead of the Adam optimizer is utilized through the classifier, and Flamingo-hyena optimization is implemented.When a test satellite image is provided to the trained BiLSTM model, then the changes in the land cover as per the satellite images will be detected.The entire research for detecting changes in the land cover will be implemented in Python.The proposed model is implemented and compared with the existing methods to reveal the effectiveness of the proposed model depending on the performance measures, such as accuracy, MSE, Precision, Recall, and F1-Score.Where:

Schematic Representation Of The Proposed Methodology
D denotes the database, and   indicates the satellite images.

PREPROCESSING
The preprocessing of the data is dispatched to eliminate the presence of undesirable fossil such as noises, which is produced through haze, clouds, and so on, which are then subjected to feature map generation, and the preprocessed data is represented by, D=S  (2)

FEATURE MAP GENERATION
The input satellite images, where the undesirable fossils get removed Where: • NIR represents the spectral reflectance or radiance in the near-infrared band of the electromagnetic spectrum.
• Red represents the spectral reflectance or radiance in the red band of the electromagnetic spectrum.

Soil-Adjusted Vegetation Index
SAVI is developed to decrease soil influences on canopy spectra by incorporating a soil adjustment factor L into the denominator of the NDVI equation, which is expressed as,

Weighted Vegetation Index (WVI)
WVI is an efficient as most of the slope-based Vegetation index (Vis).The effect of weighting the red band with the slope of the soil line is the maximization of the vegetation signal in the near-infrared band and the minimization of the effect of soil brightness, which is In this formula: • WVI represents the Weighted Vegetation Index.

Global Environmental Monitoring Index (GEMI)
It is a non-linear index to monitor global vegetation from satellites.GEMI is established for reducing the relative effects of the undesirable atmospheric perturbations present in satellite imagery while maintaining information regarding the vegetation cove, which is expressed as,

Modified Soil-Adjusted Vegetation Index (MSAVI)
The  factor present in SAVI is varying inversely to obtain the optimal adjustment ++for the soil effect.Therefore, a modified SAVI (MSAVI) replaces  factor in the SAVI expression, while reducing the influence of bare soil on SAVI, which is represented as, MSAVI = (2 * NIR + 1 -sqrt((2 * NIR + 1)^2 -8 * (NIR -Red))) / 2 (7) • NIR refers to the near-infrared reflectance value and Red refers to the red reflectance value

Modified Simple Ratio (MSR)
MSR is developed for retrieving biophysical parameters of boreal forests using remote sensing data.MSR is an advanced version of Renormalized Difference Vegetation Index (RDVI) for the purpose of linearizing connections with biophysical parameters are expressed as,
• B represents the second quantity or value.
• M is the modification factor.

FLAMINGO-HYENA OPTIMIZATION
The research utilizes Flamingo-hyena optimization for improving the classification accuracy of the BiLTSM model and assists in change detection, which is inspired through the foraging and migratory behavior of Phoenicopteridae, as well as the hunting capability of Hyaenidae.The inspiration and the optimization are briefly described here.

Inspiration
Flamingo-Hyena Optimization (FHO) algorithm is a relatively new optimization algorithm inspired by the behaviors of flamingos and hyenas in the animal kingdom.The algorithm is designed to solve optimization problems by imitating the foraging behaviors of these animals.
The flamingo-hyena optimization is established by the hybridization of the hunt optimizations such as FO [18] and SHO [19], which is processed after the hunting mechanism to portray the test result of the optimization.The FO contains a foraging behavior that simulates Phoenicopteridae, which tries to find the optimal solution in the search region, i.e. the location of the food is most abundant depending on the available limited information.

Flamingo-Hyena optimization Algorithmic Steps
The Flamingo-Hyena Optimization (FHO) algorithm is a nature-inspired optimization algorithm that imitates the behaviors of flamingos and hyenas in the animal kingdom.

Initialization:
• Initialize a population of candidate solutions, typically represented as a set of vectors or individuals.

Flamingo Behavior:
• Encourage exploration and information sharing among the population by simulating the flocking behavior of flamingos.
• Define a movement equation that determines the new position of each candidate solution based on its current position and the collective information of the population.
• The movement equation may involve factors such as the individual's current position, the best position found so far by the population, and global exploration parameters.

Hyena Behavior:
• Incorporate cooperative and competitive behaviors of hyenas for exploitation of promising regions.
• Implement interactions between candidate solutions to foster competition and cooperation.
• This can be achieved by defining fitness functions that evaluate the performance of candidate solutions and selecting individuals for competitive and cooperative interactions based on their fitness values.• Perform iterative updates by repeating the flamingo and hyena behaviors.
• Evaluate the fitness of candidate solutions based on the objective function of the optimization problem.
• Update the positions of individuals based on the movement equations and interactions defined in the flamingo and hyena behaviors.
• Continue the iterations until a termination criterion is met (e.g., a maximum number of iterations or convergence criteria).

RESULTS AND DISCUSSION
The changes in the land cover are perceived using the satellite images as well as the BiLSTM model and the accomplishment of the model is represented in this section.

EXPERIMENTAL SETUP
The FHO-BiLSTM model for Land cover change classification is implemented in Python language and the configuration included in this system is specified as Python 3.

Input Data
The Input data used in this work is real-time data obtained from the Google Pro software from 2002 to 2021 from Pune, Maharashtra at two different locations considered Dataset 1 and Dataset 2.

Accuracy:
To evaluate the accuracy of land cover change classification, a common approach is to compare the predicted classes with the ground truth or reference data.The accuracy is typically calculated as the fraction or percentage of samples that are correctly classified. Where: • Precision is the proportion of true positive predictions among all positive predictions.
• Recall is the proportion of true positive predictions among all actual positive instances.
The F1 score calculates the harmonic mean of precision and recall, providing a single value that balances the trade-off between these two metrics.The harmonic mean emphasizes lower values, meaning that the F1 score will be low if either precision or recall is low.
The F1 score ranges between 0 and 1, with 1 representing the best possible performance.
A higher F1 score indicates a better balance between precision and recall, suggesting a more accurate and reliable classification model.

Experimental analysis with Ds1
The results obtained from the input satellite image of Ds1 and the outcomes of previous approaches are illustrated in Figure 3 I. a), b), c), d), e), f), g), h), i), j), k), and l)

Experimental analysis with Ds2
The results obtained from the input satellite image of Ds2 and the outcomes of previous approaches are illustrated in Figure.

COMPARATIVE ANALYSIS OF THE BILSTM MODEL
The comparison of the FHO-BiLSTM model is determined with some traditional methods such as Unsupervised Learning (Mth1) [21], Neural Network (NN) (Mth2) [23], LSTM (Mth3) [22], BiLSTM (Mth4) [20], FSO-BiLSTM (Mth5) [18], SHO-BiLSTM (Mth6) [19].Because the input is retrieved via satellite photos, there may be a problem with clarity and dependability in some existing methods.To fix this flaw, preprocessing is done.The six vegetation index features and Resnet-101 model assist in collecting the most relevant and informative characteristics throughout the feature map generating task.The deep architecture and ability to capture minute details of ResNet-101 allow for efficient modeling and differentiation of the various land cover types.The parameters are adjusted with the aid of the suggested model in order to lower the mistake rate and boost accuracy.

Comparative analysis based on Ds1
The comparative analysis is measured based on dataset-1 and is accomplished using the data gained from the years 2002 to 2021.

BiLSTM
Classifier Model for Land Cover Change Detection and Classification ___________________________________________________________________________ Rev. Gest.Soc.Ambient.| Miami | v.18.n.5 | p.1-24 | e04962 | 2024.6 2.2 BILSTM FOR LAND COVER ANALYSIS AND LAND COVER CHANGE DETECTION To overcome major challenges, that is accuracy and reliability of change detection analyses advanced techniques, such as deep learning and Neural Networks, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), have been increasingly Source: Own development (2023).

3. 1 INPUT 8 D=S
The dataset used for the present research is two satellite images, which are collected from the varied period that are utilized as input data for the change detection in the Land cover with the interpretation formula, for Land Cover Change Detection and Classification ___________________________________________________________________________ Rev. Gest.Soc.Ambient.| Miami | v.18.n.5 | p.1-24 | e04962 | 2024. (1) further and fed forward to feature map generation.A pre-trained deep learning-based model and six vegetation index-based feature maps were executed in the feature map generation.Details of these vegetation indices are given below 3.3.1 Normalized Difference Vegetation Index NDVI is utilized for preventing unbound satellite image with medium resolution and provides limitations in certain applications, which results in being unsuitable for field observation, as well as the availability of varied factors that causes mixed natures in lowresolution satellite imagery at a plant scale, which can be expressed as, NDVI = (NIR -Red) / (NIR + Red), w1 * B1 + w2 * B2 + w3 * B3 + ... + wn * Bn) / (w1 + w2 + w3 + ... + wn) (5) ___________________________________________________________________________ Rev. Gest.Soc.Ambient.| Miami | v.18.n.5 | p.1-24 | e04962 | 2024.

7 . 6
compiled in Pycharm 2020 and implemented in Windows 10 system software.The library used is Keras, Pandas, open CV Python, Numpy, SK learn, and SK image.

4. 2
EXPERIMENTAL ANALYSIS The experimental assessment is accomplished by selecting the input satellite image by considering, a pre-trained deep learning-based model and six vegetation index-based feature maps, and the extracted features from the input images are considered as vegetation indices such as Normalized Difference Vegetation Index(NDVI), Soil-Adjusted Vegetation Index(SAVI), Weighted Vegetation Index (WVI), Global Environmental Monitoring Index(GEMI), Modified Soil-Adjusted Vegetation Index (MSAVI), and Modified Simple Ratio(MSR).

Figure 4
Figure 4 Experimental analysis for the proposed dual adaptive model for Ds2: a) input image, b) G, c)GEMI, d) MSAVI, e) MSR, f) NDVI, g) NIR, h) R, i) RVI, j) SAVI, k) WDVI, and l) Output analysis is based on dataset-1, which is gained from the year 2002(for training and testing of model) and the outcomes are evaluated by utilizing the accuracy of metrics, MSE, precision, recall, and f1 measure.It is evaluated from consecutive years from 2009 to 2021 are estimated and the outcomes gained are represented in Figure 5.The accuracy, Precision, Recall, and F1-Score rate of the proposed method is improved than the existing method and the MSE is reduced.

Figure 5 Comparative
Figure 5 Comparative Analysis using database -1 with 2009 to 2021 a) accuracy b) MSE c) Precision d) recall and e) F1-Score

Figure 6 Comparative
Figure 6 Comparative Analysis using database -2 with 2009 a) accuracy b) MSE c) Precision d) recall and e) F1-Score

Table 1
Methods for land cover classification