Advanced Deep Learning Techniques Research Articles

Plastic Waste Identification Using Deep Learning for Adequate Waste Management

Plastic waste management has emerged as a critical global challenge, prompting concerted efforts from conservation authorities and international organizations like the United Nations to enhance detection and classification strategies. This research distinguishes itself by harnessing advanced deep learning techniques to identify plastic materials at the micron level, surpassing traditional macro-level methods. The primary goal is to classify plastics into four major distinctions, addressing a key challenge in plastic segregation by accurately measuring thickness. Leveraging the YOLOv8 architecture, this approach enables precise classification of plastics into Polyethylene terephthalate (PETE), High Density Polyethylene (HDPE), low density polyethylene (LDPE0, and Polyvinyl Chloride (PVC) categories based on thickness. To achieve this, incorporation of hardware components such as ultrasonic sensors and NodeMCU for detecting thickness variations is used. By facilitating effective segregation according to environmental impact, this innovation revolutionizes waste reduction efforts, offering real-time identification and enhancing overall sustainability in plastic waste management.

Enhancing Bangladeshi License Plate Recognition: A YOLOv8 Approach with Roboflow Integration for Accuracy and Speed Optimization

Abstract: Automatic License Plate Recognition (ALPR) systems are indispensable in modern transportation services, offering crucial benefits in traffic management, parking, toll collection, and surveillance. However, implementing ALPR in Bangladesh presents challenges due to the intricacies of Bangla characters and low-resolution CCTV images. Despite ALPR's potential to enhance safety and security, issues such as license plate variations and image quality hinder its effective implementation. This research addresses these challenges by developing a customized YOLOv8 model tailored for recognizing Bengali license plates, with a focus on precise localization, character segmentation, and script deciphering. Leveraging advanced deep learning techniques, the study aims to enhance efficiency and accuracy for applications in law enforcement and traffic management. Through a comprehensive workflow integrating Roboflow and YOLOv8, effective dataset collection, annotation, augmentation, and model training are demonstrated. System evaluation on diverse test datasets confirms the reliable detection and recognition of Bangladeshi license plates, underscoring the model's practical utility and robustness in real-world scenarios. Additionally, the research showcases notable advancements in ALPR technology specific to Bangladesh's unique context. Noteworthy is the model's exceptional performance in detecting license plates from corner angles, even across various vehicle types, as well as its ability to accurately identify plates amid challenging conditions such as broken or obscured plates and low-resolution images. Moreover, the system's proficiency in scenarios with multiple license plates, like those on buses adorned with banners, contributes significantly to improved road safety and traffic management.

Micronutrient Deficiency Using Image Processing and Neural Network

Abstract: Vitamin and micronutrient deficiencies are a significant global health concern, leading to various adverse health consequences. Early detection and intervention are essential in addressing these issues. The project introduces an intelligent system that utilizes advanced deep learning techniques to identify and differentiate vitamin deficiencies in human tissue through image analysis. The approach involves an initial step of image clustering to separate and isolate problem areas from the input images. The goal is to evaluate the productivity ofimage segmentation methods, extract relevant characteristics, and compare classification results with other methods. To accomplish the objectives, a diverse dataset of facial images is gathered and preprocessed, focusing on individuals both with and without visible signs of vitamin deficiencies Following this, a CNN algorithm, inspired by models like Alex Net, is created and trained using the pre-processed dataset. The CNN isused to identify and classify features based on different types of vitamin deficiencies, enabling an automated andaccurate assessment based on facial images

MCCM: multi-scale feature extraction network for disease classification and recognition of chili leaves

Currently, foliar diseases of chili have significantly impacted both yield and quality. Despite effective advancements in deep learning techniques for the classification of chili leaf diseases, most existing classification models still face challenges in terms of accuracy and practical application in disease identification. Therefore, in this study, an optimized and enhanced convolutional neural network model named MCCM (MCSAM-ConvNeXt-MSFFM) is proposed by introducing ConvNeXt. The model incorporates a Multi-Scale Feature Fusion Module (MSFFM) aimed at better capturing disease features of various sizes and positions within the images. Moreover, adjustments are made to the positioning, activation functions, and normalization operations of the MSFFM module to further optimize the overall model. Additionally, a proposed Mixed Channel Spatial Attention Mechanism (MCSAM) strengthens the correlation between non-local channels and spatial features, enhancing the model’s extraction of fundamental characteristics of chili leaf diseases. During the training process, pre-trained weights are obtained from the Plant Village dataset using transfer learning to accelerate the model’s convergence. Regarding model evaluation, the MCCM model is compared with existing CNN models (Vgg16, ResNet34, GoogLeNet, MobileNetV2, ShuffleNet, EfficientNetV2, ConvNeXt), and Swin-Transformer. The results demonstrate that the MCCM model achieves average improvements of 3.38%, 2.62%, 2.48%, and 2.53% in accuracy, precision, recall, and F1 score, respectively. Particularly noteworthy is that compared to the original ConvNeXt model, the MCCM model exhibits significant enhancements across all performance metrics. Furthermore, classification experiments conducted on rice and maize disease datasets showcase the MCCM model’s strong generalization performance. Finally, in terms of application, a chili leaf disease classification website is successfully developed using the Flask framework. This website accurately identifies uploaded chili leaf disease images, demonstrating the practical utility of the model.

Handwritten Based Alzheimer Disease Prediction from One Dimensional Datasets Using Deep Learning

Because to their high cost, sensitivity, and difficulty in completing surgeries, brain-related disorders are among the most challenging conditions. On the other hand, since the procedure's outcomes could be negative, the operation itself does not have to succeed. Alzheimer's disease, which affects adults and causes varied degrees of memory loss and knowledge forgetfulness, is one of the most prevalent brain diseases. based on the state of each patient. For these reasons, using user-handwritten datasets to categorise memory loss and determine the patient's evaluation of Alzheimer's disease at every given level is crucial. This work offers a novel method for predicting Alzheimer's disease by using advanced deep learning techniques on handwritten data. Alzheimer's is a degenerative. Alzheimer's disease, a neurological condition that progresses and necessitates prompt diagnosis and appropriate treatment. Traditional diagnostic techniques are mostly based on clinical evaluations and imaging, which are frequently inaccessible and expensive. This study investigates the unrealized potential of handwritten data as a special kind of Alzheimer's disease prediction. The dataset provides a broad picture of cognitive impairments by containing handwritten samples collected from individuals in varying states of cognition. Using deep learning architectures, such as the multi-layer perceptron method, the suggested model takes use of the temporal dependencies found in sequential handwritten patterns. These designs show promise for ensuring the temporal characteristics of handwritten data are captured with subtle features. The handwritten input is converted into a format suitable for deep learning using feature extraction techniques, which helps with efficient model training. A thorough assessment of the model's performance is conducted using common metrics including specificity, sensitivity, and accuracy. The goal is to determine whether the model can correctly forecast Alzheimer's disease based just on the unique features present in handwritten samples.

Utilizing portable electroencephalography to screen for pathology of Alzheimer’s disease: a methodological advancement in diagnosis of neurodegenerative diseases

BackgroundThe current biomarker-supported diagnosis of Alzheimer’s disease (AD) is hindered by invasiveness and cost issues. This study aimed to address these challenges by utilizing portable electroencephalography (EEG). We propose a novel, non-invasive, and cost-effective method for identifying AD, using a sample of patients with biomarker-verified AD, to facilitate early and accessible disease screening.MethodsThis study included 35 patients with biomarker-verified AD, confirmed via cerebrospinal fluid sampling, and 35 age- and sex-balanced healthy volunteers (HVs). All participants underwent portable EEG recordings, focusing on 2-minute resting-state EEG epochs with closed eyes state. EEG recordings were transformed into scalogram images, which were analyzed using “vision Transformer(ViT),” a cutting-edge deep learning model, to differentiate patients from HVs.ResultsThe application of ViT to the scalogram images derived from portable EEG data demonstrated a significant capability to distinguish between patients with biomarker-verified AD and HVs. The method achieved an accuracy of 73%, with an area under the receiver operating characteristic curve of 0.80, indicating robust performance in identifying AD pathology using neurophysiological measures.ConclusionsOur findings highlight the potential of portable EEG combined with advanced deep learning techniques as a transformative tool for screening of biomarker-verified AD. This study not only contributes to the neurophysiological understanding of AD but also opens new avenues for the development of accessible and non-invasive diagnostic methods. The proposed approach paves the way for future clinical applications, offering a promising solution to the limitations of advanced diagnostic practices for dementia.

Deep learning based automated estimation of urban green space index from satellite image: A case study

The green area factor model is a crucial tool for conserving and creating urban greenery and ecosystem services within neighborhood land. This model serves as a valuable index, streamlining the planning, assessment, and comparison of local-scale green infrastructures. However, conventional on-site measurements of the green area factor are resource intensive. In response, this study pioneers a computational approach that integrates ecological and social dimensions to estimate the green area factor. Employing satellite remote sensing and advanced deep learning techniques, the methodology utilizes satellite orthophotos of urban areas subjected to semantic segmentation, identifying and categorizing diverse green elements. Ground truths are established through on-site measurements of green area factors and satellite orthophotos from benchmarking sites in Örebro, Sweden. Results reveal an 82.0% average F1-score for semantic segmentations, signifying a favourable correlation between computationally estimated and measured green area factors. The proposed methodology is potential for adapting to various urban settings. In essence, this research introduces a promising, cost-effective solution for assessing urban greenness, particularly beneficial for urban administrators and planners aiming for insightful and comprehensive green strategies in city planning.

Predicting atrial fibrillation ablation success: a novel approach using pre-procedure sinus rhythm ecg and deep learning

Abstract Introduction Pulmonary vein isolation (PVI) has been the corner stone of the treatment of patients with atrial fibrillation (AF), yet its efficacy varies notably between patient groups. Specifically, it shows reduced efficacy in individuals with persistent AF and those with atrial fibrosis. Recent advancements suggest that additional substrate modification could enhance outcomes in specific patient groups. Therefore, predicting the outcomes of lone PVI could help in stratifying patients who would benefit most from this additional substrate modification. Addressing this gap, our study explores the prognostication of PVI outcomes using pre-procedural 12-lead ECG recordings taken in Sinus rhythm, employing advanced deep learning techniques. Methods This retrospective study analyzed 209 patients with paroxysmal and persistent AF who underwent PVI. These patients were followed up for 12 months after the blanking period. Pre-procedural sinus rhythm 12-lead ECGs were processed using an innovative deep learning architecture, integrating Long Short-Term Memory and Convolutional Neural Networks with a cross-attention module. This multimodal deep learning model was trained using ECG and clinical covariates like age, sex, and BMI to predict binary outcomes – AF type (Paroxysmal or Persistent) and AF recurrence. The model's efficacy was evaluated through a 5-fold stratified cross-validation to ensure accuracy and reliability. Model was also tested using only ECG data. Results The multimodal deep learning model performed significantly better as compared to the model trained using ECG data only. The multimodal deep learning model achieved an accuracy of 75.56% and a sensitivity of 82% in prognosticating ablation outcomes. Additionally, the model demonstrated a significant capability in classifying AF into paroxysmal or persistent types with an accuracy of 86.67% and a recall of 89%. Conclusion Our deep learning approach in analyzing sinus rhythm ECG recordings shows promising results in predicting PVI outcomes and identifying AF types. This approach could offer a more accessible and cost-effective tool to stratify AF patients undergoing PVI, identifying those less likely to benefit from standard procedures. Such insights can assist physicians and patients in determining ablation candidacy and selecting the most appropriate ablation strategy and technique.Proposed StrategyDeep learning model approach

Integrating Yolo V5 Analysis and KNN to Improve Lung Cancer Detection

Lung cancer is a leading cause of cancer-related mortality globally, emphasizing the urgent need for early detection and accurate diagnosis. This project aims to leverage advanced deep learning techniques, specifically YOLO-v5 (You Only Look Once) for object detection, and the k-Nearest Neighbors (kNN) algorithm for unsupervised learning, to enhance the detection and analysis of lung cancer from CT scan images. YOLO-v5, known for its exceptional speed and accuracy in detecting objects within images, will be used to identify and localize lung nodules, which are potential indicators of lung cancer. Simultaneously, we will employ the kNN algorithm in a novel application of unsupervised learning to cluster CT scan images based on the similarity of detected lung tumors, enabling the identification of patterns and characteristics that may correlate with specific types of lung cancer. This project involves collecting and preprocessing a diverse dataset of CT images annotated with radiologist insights to train the YOLO-v5 model. Subsequently, the kNN algorithm will be applied to perform clustering on the detected tumors. By achieving high accuracy in nodule detection and effectively clustering similar tumors, the system aims to become an invaluable tool for radiologists, providing rapid diagnostic assistance and facilitating a deeper understanding of lung cancer characteristics.

Exposing Deep Fake Face Detection using LSTM and CNN

The rapid advancement of deep learning techniques, creating realistic multimedia content has become increasingly accessible, leading to the proliferation of DeepFake technology. DeepFake utilizes generative deep learning algorithms to produce or modify face features in a highly realistic manner, often making it challenging to differentiate between real and manipulated media. This technology, while beneficial in fields such as entertainment and education, also poses significant threats, including misinformation and identity theft. Consequently, detecting DeepFakes has become a critical area of research. In this paper, we propose a novel approach to DeepFake face detection by integrating Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) networks. Our method leverages the strengths of CNNs in spatial feature extraction and LSTMs in temporal sequence modeling to enhance detection accuracy. The CNN component captures intricate facial features, while the LSTM analyzes the temporal dynamics of video frames. We evaluate our model on several benchmark datasets, including Celeb-DF (v2), DeepFake Detection Challenge Preview, and FaceForensics++. Experimental results demonstrate that our hybrid CNN-LSTM model achieves state-of-the-art performance, surpassing existing methods in both accuracy and robustness. This study highlights the potential of combining CNN and LSTM architectures for effective DeepFake detection, contributing to the ongoing efforts to safeguard against digital media manipulation

A comprehensive review of deep learning in EEG-based emotion recognition: classifications, trends, and practical implications

Emotion recognition utilizing EEG signals has emerged as a pivotal component of human–computer interaction. In recent years, with the relentless advancement of deep learning techniques, using deep learning for analyzing EEG signals has assumed a prominent role in emotion recognition. Applying deep learning in the context of EEG-based emotion recognition carries profound practical implications. Although many model approaches and some review articles have scrutinized this domain, they have yet to undergo a comprehensive and precise classification and summarization process. The existing classifications are somewhat coarse, with insufficient attention given to the potential applications within this domain. Therefore, this article systematically classifies recent developments in EEG-based emotion recognition, providing researchers with a lucid understanding of this field’s various trajectories and methodologies. Additionally, it elucidates why distinct directions necessitate distinct modeling approaches. In conclusion, this article synthesizes and dissects the practical significance of EEG signals in emotion recognition, emphasizing its promising avenues for future application.

Robust Forest Fire Detection Method for Surveillance Systems Based on You Only Look Once Version 8 and Transfer Learning Approaches

Forest fires have emerged as a significant global concern, exacerbated by both global warming and the expanding human population. Several adverse outcomes can result from this, including climatic shifts and greenhouse effects. The ramifications of fire incidents extend widely, impacting human communities, financial resources, the natural environment, and global warming. Therefore, timely fire detection is essential for quick and effective response and not to endanger forest resources, animal life, and the human economy. This study introduces a forest fire detection approach utilizing transfer learning with the YOLOv8 (You Only Look Once version 8) pretraining model and the TranSDet model, which integrates an improved deep learning algorithm. Transfer Learning based on pre-trained YoloV8 enhances a fast and accurate object detection aggregate with the TranSDet structure to detect small fires. Furthermore, to train the model, we collected 5200 images and performed augmentation techniques for data, such as rotation, scaling, and changing due and saturation. Small fires can be detected from a distance by our suggested model both during the day and at night. Objects with similarities can lead to false predictions. However, the dataset augmentation technique reduces the feasibility. The experimental results prove that our proposed model can successfully achieve 98% accuracy to minimize catastrophic incidents. In recent years, the advancement of deep learning techniques has enhanced safety and secure environments. Lastly, we conducted a comparative analysis of our method’s performance based on widely used evaluation metrics to validate the achieved results.

Age-Invariant Cross-Age Face Verification using Transfer Learning

The integration of face verification technology has become indispensable in numerous safety and security software systems. Despite its promising results, the field of face verification encounters significant challenges due to age-related disparities. Human facial characteristics undergo substantial transformations over time, leading to diverse variations including changes in facial texture, morphology, facial hair, and eyeglass adoption. This study presents a pioneering methodology for cross-age face verification, utilizing advanced deep learning techniques to extract resilient and distinctive facial features that are less susceptible to age-related fluctuations. The feature extraction process combines handcrafted features like Local Binary Pattern/Histogram of Oriented Gradients with deep features from MobileNetV2 and VGG-16 networks. As the texture of the facial skin defines the age related characteristic the well-known texture feature extractors like LBP and HoG is preferred. These features are concatenated to achieve fusion, and subsequent layers fine-tune them. Experimental validation utilizing the Cross-Age Celebrity Dataset demonstrates remarkable efficacy, achieving an accuracy of 98.32%.

Field Validation of Deep-Learning-Based Ground Penetrating Radar Image Analysis for Advancing Subsurface Distress Detection

This research introduces an innovative method for detecting subsurface cracks within pavements by leveraging ground penetrating radar (GPR) technology in conjunction with advanced deep learning techniques. Its primary aim is to significantly improve the accuracy and efficiency of pavement assessment, particularly for operational and maintenance purposes. The proposed model, GPR-YOLOR (You Only Learn One Representation), extends the YOLOR framework and incorporates a region of interest within the top pavement layer to detect subsurface cracks. While the model can be trained with annotated data, the main challenge lies in validating results in the field because of the inability to visually inspect subsurface conditions and the high cost associated with direct coring. To overcome this challenge, we propose an alternative approach that utilizes the co-occurrence of surface cracks as pseudo labels, allowing for easy verification. To ensure that surface cracks correspond to subsurface cracks, the focus is exclusively on transverse cracks that develop in a bottom-up manner, such as fatigue and reflective cracks. Through this methodology, our GPR-YOLOR model achieves an F1 score of 0.72, with a precision of 0.76 and a recall of 0.68. The results from field validation underscore the effectiveness of the GPR-YOLOR model in accurately identifying subsurface cracks, highlighting its practical significance in conducting field condition assessments.

A bizarre synthesized cascaded optimized predictor (BizSCOP) model for enhancing security in cloud systems

Due to growing network data dissemination in cloud, the elasticity, pay as you go options, globally accessible facilities, and security of networks have become increasingly important in today's world. Cloud service providers, including AWS, Azure, GCP, and others, facilitate worldwide expansion within minutes by offering decentralized communication network functions, hence providing security to cloud is still remains a challenging task. This paper aims to introduce and evaluate the Biz-SCOP model, a novel intrusion detection system developed for cloud security. The research addresses the pressing need for effective intrusion detection in cloud environments by combining hybrid optimization techniques and advanced deep learning methodologies. The study employs prominent intrusion datasets, including CSE-CIC-IDS 2018, CIC-IDS 2017, and a cloud intrusion dataset, to assess the proposed model's performance. The study's design involves implementing the Biz-SCOP model using Matlab 2019 software on a Windows 10 OS platform, utilizing 8 GB RAM and an Intel core i3 processor. The hybrid optimization approach, termed HyPSM, is employed for feature selection, enhancing the model's efficiency. Additionally, an intelligent deep learning model, C2AE, is introduced to discern friendly and hostile communication, contributing to accurate intrusion detection. Key findings indicate that the Biz-SCOP model outperforms existing intrusion detection systems, achieving notable accuracy (99.8%), precision (99.7%), F1-score (99.8%), and GEO (99.9%). The model excels in identifying various attack types, as demonstrated by robust ROC analysis. Interpretations and conclusions emphasize the significance of hybrid optimization and advanced deep learning techniques in enhancing intrusion detection system performance. The proposed model exhibits lower computational load, reduced false positives, ease of implementation, and improved accuracy, positioning it as a promising solution for cloud security.

Admission Prediction Using Time Series Analysis

Efficient resource allocation and strategic planning in educational institutions heavily rely on accurate predictions of student admissions. This paper presents a detailed investigation into the application of time series analysis techniques for admission prediction. We explore the utilization of historical admission data, spanning multiple academic cycles, to train and evaluate several time series models.The study encompasses a comparative analysis of traditional statistical models such as autoregressive integrated moving average (ARIMA) and seasonal autoregressive integrated moving average (SARIMA) alongside more advanced deep learning techniques like long short-term memory (LSTM) networks. By evaluating the performance metrics, including accuracy, robustness, and computational efficiency, we aim to identify the most suitable model for admission prediction tasks. Moreover, the research delves into the impact of various external factors such as changes in the academic calendar, socio-economic indicators, and demographic shifts on admission patterns. Understanding these factors is crucial for enhancing the predictive capabilities of the models and enabling institutions to adapt their strategies accordingly. The experimental results and comparative analysis provide valuable insights for educational institutions, enabling them to make data-driven decisions regarding enrollment management strategies and resource allocation. Ultimately, this research contributes to the advancement of admission prediction methodologies, facilitating more efficient and informed decision-making processes in the educational domain.

Enhancing Emergency Vehicle Detection: A Deep Learning Approach with Multimodal Fusion

Emergency vehicle detection plays a critical role in ensuring timely responses and reducing accidents in modern urban environments. However, traditional methods that rely solely on visual cues face challenges, particularly in adverse conditions. The objective of this research is to enhance emergency vehicle detection by leveraging the synergies between acoustic and visual information. By incorporating advanced deep learning techniques for both acoustic and visual data, our aim is to significantly improve the accuracy and response times. To achieve this goal, we developed an attention-based temporal spectrum network (ATSN) with an attention mechanism specifically designed for ambulance siren sound detection. In parallel, we enhanced visual detection tasks by implementing a Multi-Level Spatial Fusion YOLO (MLSF-YOLO) architecture. To combine the acoustic and visual information effectively, we employed a stacking ensemble learning technique, creating a robust framework for emergency vehicle detection. This approach capitalizes on the strengths of both modalities, allowing for a comprehensive analysis that surpasses existing methods. Through our research, we achieved remarkable results, including a misdetection rate of only 3.81% and an accuracy of 96.19% when applied to visual data containing emergency vehicles. These findings represent significant progress in real-world applications, demonstrating the effectiveness of our approach in improving emergency vehicle detection systems.

STABLE DIFFUSION TEXT TO IMAGE USING AI

The Stable Diffusion Text-to-Image Generation Project is an innovative endeavor in the field of generative adversarial networks (GANs) and natural language processing (NLP). This project aims to bridge the semantic gap between textual descriptions and visual content by utilizing the Stable Diffusion training framework to generate highly realistic and coherent images from text prompts. The project leverages recent advancements in deep learning techniques to tackle the challenging task of text-to image synthesis. The project introduces an innovative approach at the crossroads of generative adversarial networks (GANs) and natural language processing (NLP). It aims to bridge the semantic gap between textual descriptions and visual content by utilizing the Stable Diffusion training framework. The key goal is to generate highly realistic and coherent images from text prompts, leveraging recent deep learning Stable Diffusion. The Stable Diffusion training framework plays a central role in this project. It’s a sophisticated training methodology for GANs, designed to stabilize the training process. GANs have exhibited great potential in generating images but often suffer from issues like mode collapse and training instability. Keywords: Stable Diffusion, Text-to-Image Generation, Image Synthesis, Natural Language Processing(NLP), Generative Adversarial Networks (GANs)

Surface Defect Detection of Aluminum Profiles Based on Multiscale and Self-Attention Mechanisms.

To address the various challenges in aluminum surface defect detection, such as multiscale intricacies, sensitivity to lighting variations, occlusion, and noise, this study proposes the AluDef-ClassNet model. Firstly, a Gaussian difference pyramid is utilized to capture multiscale image features. Secondly, a self-attention mechanism is introduced to enhance feature representation. Additionally, an improved residual network structure incorporating dilated convolutions is adopted to increase the receptive field, thereby enhancing the network's ability to learn from extensive information. A small-scale dataset of high-quality aluminum surface defect images is acquired using a CCD camera. To better tackle the challenges in surface defect detection, advanced deep learning techniques and data augmentation strategies are employed. To address the difficulty of data labeling, a transfer learning approach based on fine-tuning is utilized, leveraging prior knowledge to enhance the efficiency and accuracy of model training. In dataset testing, our model achieved a classification accuracy of 97.6%, demonstrating significant advantages over other classification models.

Generalizable data driven full waveform inversion for complex structures and severe topographies

Traditionally, simplification has been used in scientific modeling practices. However, recent advancements in deep learning techniques have provided a means to represent complex models. As a result, deep neural networks should be able to approximate the complex models, with a high degree of generalization.To achieve generalization, it is necessary to have a diverse range of examples in the training of the neural network, for example in data-driven FWI, training data should cover the expected subsurface models. To meet this requirement, we porposed a method to creat geologically meaningful velocity models with complex structures and severe topography. However, it is important to note that generalization comes with its own set of challenges.Because of significant variation in topography of the generated velocity models, we need to include this information as an additional input data in training of the network. Therefore, we have transformed the seismic data to a fixed datum to incorporate geometric information. Additionally, we have enhanced the network's performance by introducing a term in the network loss function. Multiple metrics have been employed to evaluate the performance of the network. The results indicate that by providing the necessary information to the network and employing computational techniques to refine the model's accuracy, deep neural networks are capable of accurately estimating velocity models in complex environments characterized by extreme topography.