Deep Belief Network Model Research Articles

An Intelligent Facial Expression Recognition System Using a Hybrid Deep Convolutional Neural Network for Multimedia Applications

Recognizing facial expressions plays a crucial role in various multimedia applications, such as human–computer interactions and the functioning of autonomous vehicles. This paper introduces a hybrid feature extraction network model to bolster the discriminative capacity of emotional features for multimedia applications. The proposed model comprises a convolutional neural network (CNN) and deep belief network (DBN) series. First, a spatial CNN network processed static facial images, followed by a temporal CNN network. The CNNs were fine-tuned based on facial expression recognition (FER) datasets. A deep belief network (DBN) model was then applied to integrate the segment-level spatial and temporal features. Deep fusion networks were jointly used to learn spatiotemporal features for discrimination purposes. Due to its generalization capabilities, we used a multi-class support vector machine classifier to classify the seven basic emotions in the proposed model. The proposed model exhibited 98.14% recognition performance for the JaFFE database, 95.29% for the KDEF database, and 98.86% for the RaFD database. It is shown that the proposed method is effective for all three databases, compared with the previous schemes for JAFFE, KDEF, and RaFD databases.

Mountain Gazelle Optimizer with Deep Learning Driven Satirical News Classification on Low-resource Language Corpus

The development of satirical and fake news on digital platforms has source of major concern about the spread of misinformation and its control on society. As part of the Arabic language, fake news detection (FND) shows particular problems because of language difficulties and the scarcity of labeled data. FND on Arabic corpus utilizing deep learning (DL) contains leveraging advanced neural network (NN) techniques and methods to automatically recognize and classify deceptive data in the Arabic language text. This procedure is vital in combating the spread of disinformation and misinformation, promoting media literacy, and make sure the credibility of data sources for the Arabic-speaking community. Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs) are common selections for FND because of their capability for learning hierarchical features and model sequential data from the text. In this view, this study develops a Mountain Gazelle Optimizer with Deep Learning-Driven Fake News Classification on Arabic Corpus (MGODL-FNCAC) technique. The presented MGODL-FNCAC approach aims to increase the performance of the fake news classification on the Arabic corpus. Primarily, the MGODL-FNCAC technique involves different stages of pre-processing to make the input data compatible for classification. For fake news detection, the MGODL-FNCAC technique applies the deep belief network (DBN) model. At last, the MGO approach can be used for the better hyperparameter tuning of the DBN approach, which supports in enhancing the overall training process and detection rate. The simulation outcomes of the MGODL-FNCAC technique can be examined on Arabic corpus data. The extensive outcomes exhibit the importance of the MGODL-FNCAC system over other methodologies with maximum accuracy of 97.68% and 95.14% on Covid19Fakes and Satirical dataset, respectively.

The prediction of sleep quality using wearable-assisted smart health monitoring systems based on statistical data

The technology, which plays a significant role in our lives, has made it possible for many of the appliances and gadgets we use on a daily basis to be monitored and controlled remotely. Health and fitness data is collected by wearable devices attached to patients' bodies. A number of parties could benefit from this technology, including doctors, insurers, and health providers. This technology, including smartwatches, smart ring, smart cloth wristbands, and GPS shoes, is frequently used for fitness and wellness since it allows users to track their day-to-day health. Devices that compute the sleep characteristics by storing sleep movements fall within the category of wearables worn on the wrist. In order to lead a healthy lifestyle, sleep is crucial. Inadequate sleep can harm one's physical, mental, and emotional well-being and increase the risk of developing a number of ailments, including stress, heart disease, high blood pressure, insulin resistance, and other conditions. Deep learning (DL) models have recently been used to forecast sleep-quality based on wearables information from the awake hours. Deep learning has been demonstrated to be capable of predicting sleep efficiency based on wearable data obtained during awake periods. In this regard, this study creates a novel deep learning model for wearables-enabled smart health monitoring system (DLM-WESHMS) for the prediction of sleep quality. The wearables are initially able to collect data linked to sleep-activity using the described DLM-WESHMS approach. The data is then put through pre-processing to create a standard format. Using the DLM-WESHMS, sleep quality is predicted using the deep belief network (DBN) model. The DBN model uses the auto-encoders algorithm (AEA) to predict popularity, which improves the accuracy of its predictions of sleep quality. The experimental outcomes of the DLM-WESHMS approach are investigated using several metrics. The DLM-WESHMS model performs significantly better than other models, according to a thorough comparison analysis.

Energy aware routing with optimal deep learning based anomaly detection in 6G-IoT networks

The sixth generation (6G) wireless communication networks are being planned for the transformation of client services and applications through the Internet of Things (IoT) technology that is headed towards a completely intelligent and autonomous system. The IoT-6G network offers ubiquitous applications on the basis of wireless communication technologies. However, various problems need to be critically addressed in order to realize this technological application. Primarily, the devices encounter energy constraints that prevent it from smart functioning due to limited battery power; so the connectivity issue arises resulting in the failure of the connections, once the device’s energy gets exhausted. Besides, security also remains a challenging issue which can be solved with the help of Machine Learning (ML) and Deep Learning (DL)-based anomaly detection techniques. In this background, the current study presents a Metaheuristic Energy-Aware Routing with Optimal DL-based Anomaly Detection Technique (MER-ODLADT) for 6G-IoT networks. The presented MER-ODLADT technique encompasses two major processes namely, routing and anomaly detection. For the routing process, the proposed MER-ODLADT technique applies the Marine Predator’s Optimization (MPO) algorithm with a fitness function involving multiple input parameters. On the other hand, the MER-ODLADT technique employs the Eco-Geography Optimization with Deep Belief Network (EGODBN) model for anomaly detection and classification. The performance of the proposed MER-ODLADT technique was validated using a benchmark dataset and the results were examined under distinct measures. The experimental outcomes demonstrate the enhanced performance of the proposed MER-ODLADT technique over other existing approaches with a maximum sensitivity of 98.34 %, specificity of 97.91 %, accuracy of 99.43 %, precision of 99.58 % and an F1-score of 98.41 %.

Continuous Motion Estimation of Knee Joint Based on a Parameter Self-Updating Mechanism Model.

Estimation of continuous motion of human joints using surface electromyography (sEMG) signals has a critical part to play in intelligent rehabilitation. Traditional methods always use sEMG signals as inputs to build regression or biomechanical models to estimate continuous joint motion variables. However, it is challenging to accurately estimate continuous joint motion in new subjects due to the non-stationarity and individual differences in sEMG signals, which greatly limits the generalisability of the method. In this paper, a continuous motion estimation model for the human knee joint with a parameter self-updating mechanism based on the fusion of particle swarm optimization (PSO) and deep belief network (DBN) is proposed. According to the original sEMG signals of different subjects, the method adaptively optimized the parameters of the DBN model and completed the optimal reconstruction of signal feature structure in high-dimensional space to achieve the optimal estimation of continuous joint motion. Extensive experiments were conducted on knee joint motions. The results suggested that the average root mean square errors (RMSEs) of the proposed method were 9.42° and 7.36°, respectively, which was better than the results obtained by common neural networks. This finding lays a foundation for the human-robot interaction (HRI) of the exoskeleton robots based on the sEMG signals.

Research on a small sample feature transfer method for fault diagnosis of reciprocating compressors

The assumption of a data-driven fault diagnosis model is that both the training data and the test data come from the same probability distribution. In the diagnosis of reciprocating compressors, problems such as sensor failures and changes in installation locations often occur, making it difficult to obtain effective, stable and sufficient data. The sample data is not sufficient to meet the requirements of the fault diagnosis model. We propose a fault diagnosis method for reciprocating compressors based on feature transfer at small sample sizes. First, the sample data is projected to the Regenerated Kernel Hilbert Space (RKHS) using Transfer Component Analysis (TCA). Second, the minimization of distributional differences across domains is measured by the maximum mean difference (MMD). Then, combined with the adaptive feature extraction capability of Deep Belief Network (DBN), the already labeled source domain and unlabeled target domain data are aggregated during the training phase to diagnose faults. This approach uses a novel optimized learning approach, free energy in persistent contrastive divergence, in DBN learning and training. It solves the problem of DBN classification ability degradation in long-term training. Finally, a low-power single-acting reciprocating compressor experimental platform is built to carry out fault diagnosis experiments. The research results show that better diagnostic accuracy can be obtained by reducing the distribution difference of source and target domain data in projected high-dimensional RKHS through the common transfer components learned by TCA. Compared with traditional small-sample machine learning algorithms and DBN models, this approach has strong generalization ability, and the accuracy rate is as high as 92%. This study proves the superiority of both transfer learning (TL) and deep learning fault diagnosis models in solving the problem of less fault data and provides a theoretical reference for accurate fault diagnosis of complex dynamic equipment in the case of small samples.

A landslide susceptibility assessment method based on auto-encoder improved deep belief network

Abstract The landslide susceptibility assessment is an essential part of landslide disaster risk identification and prevention. However, the binarization of the hidden layer limits the parameterization ability of the conditional probability of visible layer, making the training process of restricted Boltzmann machine more difficult and further limiting the accuracy and efficiency of deep belief network (DBN) model in landslide susceptibility assessment. Therefore, this study proposed a landslide susceptibility assessment method based on Auto-Encoder (AE)-modified DBN. Zhouqu County, Gansu Province in the People’s Republic of China, was selected as the study area. Historical landslides in Zhouqu County were identified using small baseline subset interferometric synthetic aperture radar technology and optical image. Landslide factors were screened based on a geographical detector and stepwise regression method. The Logcosh loss function and determinant coefficient R 2 index were used to evaluate the training process of the AE model, and the balanced cross entropy loss function was used to evaluate the entire network training process. In addition, the area under the curve (AUC) of the synthetical index model (SIM), support vector machine (SVM), and multilayer perceptron (MLP) were compared and evaluated. The results indicated that the proposed model could significantly improve the accuracy of landslide susceptibility assessment. The AUC value of the proposed model was 0.31, 0.12, and 0.11 higher than that of SIM, SVM, and MLP, respectively. Therefore, the improved DBN model based on AE proposed is reliable for early landslide identification and prediction.

Modified rat swarm optimization with deep learning model for robust recycling object detection and classification

Biomass residues consist of sewage effluents and sludges, crop wastes, non-recyclable municipal solid waste industrial and domestic greywater, and much more. Recycling is regarded as one such significant disposal technique. Considering the intelligent classification and detection of solid waste as an essential factor in consumption and recycling, multi-object solid waste identification and classification methodology on the basis of transfer learning (TL) are presented. This study develops a Modified Rat Swarm Optimization with Deep Learning for Robust Recycling Object Detection and Classification (MRSODL-RODC) model. Primarily, fully convolutional network (FCN) is applied for the identification of waste objects. Moreover, MRSO with deep belief network (DBN) model is applied for object detection process. The design of the MRSO algorithm for the DBN technique, showing the novelty of our work. The performance validation of the MRSODL-RODC model can be executed with the help of the benchmark dataset from Kaggle repository. The experimental outcomes demonstrated the better performance of the MRSODL-RODC model over recent approaches with higher accuracy of 99.20%.

Optimized deep belief network and unsupervised deep learning methods for disease prediction

Due to the vast amount of patient health data, automated healthcare systems still struggle to classify and diagnose various ailments. Learning redundant data also reduces categorization accuracy. A Deep Belief Network (DBN) has been used to precisely extract the most important aspects from clinical data by ignoring irrelevant/redundant features. Due of many learning variables, training is complicated. Similarly, the hybrid model has been employed by ensemble Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) and Support Vector Machine (SVM) to categorize diseases. But, its efficiency depends on the proper choice of kernels and hyper-parameters. Therefore, this paper develops an efficient feature extraction and classification model for healthcare systems. First, several medical data related to the patient’s health are collected. Then, an Optimized DBN (ODBN) model is presented for maximizing the accurateness of DBN by optimizing the learning variables depends on the Ant Lion Optimization (ALO) algorithm. With learning ODBN, the most relevant features are extracted with reduced computational complexity. After that, the CNN-LSTM with Unsupervised Fine-tuned Deep Self-Organizing Map (UFDSOM)-based classifier model is designed to categorize the extracted features into categories of illnesses. In this novel classifier, dropout normalization and parameter tuning processes are applied to avoid overfitting and optimize the hyper-parameters, which results in a less training period. In the end, studies utilizing publically accessible datasets show that the ODBN with CNN-LSTM-UFDSOM system outperforms classical models by 98.23%.

Combined Tri-Classifiers for IoT Botnet Detection with Tuned Training Weights

Although IoT sectors seem more popular and pervasively, they struggle with hazards. The botnet is one of the largest security dangers associated with IoT. It enables malicious software to administer and attack private network equipment collectively without the owners’ knowledge. Although many studies have used ML to detect botnets, these are either not very effective or only work with specific types of botnets or devices. As a result, the detection model for deep learning ideas is the focus of this research. It entails three key processes: (a) preprocessing, (b) feature extraction, and (c) classification. The input data are initially preprocessed using an improved data normalization approach. The preprocessed data are used to extract a number of features, including Tanimoto coefficient features, improved differential holoentropy-based features, Pearson r correlation-based features, and others. The detection process will be completed by an ensemble classification model that randomly shuffles models like the Deep Belief Network (DBN) model, Bidirectional Gated Recurrent Unit (Bi-GRU), and Long Short-Term Memory (LSTM). Bi-GRU, DBN, and LSTM will be averaged to provide the ensemble results. Bi-GRU is trained using the Self Improved Blue Monkey Optimization (SIBMO) Algorithm by selecting the optimal weights, which increases the detection accuracy. The overall performance of the suggested work is then evaluated in relation to other existing models using various methodologies. In comparison to existing methods, the created ensemble classifier [Formula: see text] SIBMO scheme obtains the highest accuracy (93%) at a learning percentage of 90%.

Optimal design of cold-formed steel face-to-face built-up columns through deep belief network and genetic algorithm

In this paper, a machine-learning optimisation framework for cold-formed steel (CFS) face-to-face built-up columns was proposed using Deep Belief Network (DBN) and Genetic Algorithm (GA). Firstly, an elasto-plastic non-linear finite element (FE) model was developed and validated against the available test results in the literature. Based on the validated FE model, a total of 1,100 datasets were generated, which were then used to train the DBN model. As the DBN model was proven accurate at predicting the axial strength of CFS face-to-face built-up columns, the trained DBN model was therefore used in the optimisation procedure. Through the optimisation procedure, considerable increases of up to 184.7% in axial strength were acquired for the optimised sections, when compared to the axial strength of initial CFS box section, which is commonly used in the industry. Furthermore, the optimised results obtained from the trained DBN model were compared with the results predicted by the Direct Strength Method (DSM). It was found that the DBN-GA optimised results acquired using FE datasets outperformed the results predicted by the DSM.

Retracted: Analysis of Distributed English Chunk Recognition under Correlation Evaluation Based on Deep Belief Network Model

Retracted: Analysis of Distributed English Chunk Recognition under Correlation Evaluation Based on Deep Belief Network Model

Anomaly Detection in Pedestrian Walkways for Intelligent Transportation System Using Federated Learning and Harris Hawks Optimizer on Remote Sensing Images

Anomaly detection in pedestrian walkways is a vital research area that uses remote sensing, which helps to optimize pedestrian traffic and enhance flow to improve pedestrian safety in intelligent transportation systems (ITS). Engineers and researchers can formulate more potential techniques and tools with the power of computer vision (CV) and machine learning (ML) for mitigating potential safety hazards and identifying anomalies (i.e., vehicles) in pedestrian walkways. The real-world challenges of scenes and dynamics of environmental complexity cannot be handled by the conventional offline learning-based vehicle detection method and shallow approach. With recent advances in deep learning (DL) and ML areas, authors have found that the image detection issue ought to be devised as a two-class classification problem. Therefore, this study presents an Anomaly Detection in Pedestrian Walkways for Intelligent Transportation Systems using Federated Learning and Harris Hawks Optimizer (ADPW-FLHHO) algorithm on remote sensing images. The presented ADPW-FLHHO technique focuses on the identification and classification of anomalies, i.e., vehicles in the pedestrian walkways. To accomplish this, the ADPW-FLHHO technique uses the HybridNet model for feature vector generation. In addition, the HHO approach is implemented for the optimal hyperparameter tuning process. For anomaly detection, the ADPW-FLHHO technique uses a multi deep belief network (MDBN) model. The experimental results illustrated the promising performance of the ADPW-FLHHO technique over existing models with a maximum AUC score of 99.36%, 99.19%, and 98.90% on the University of California San Diego (UCSD) Ped1, UCSD Ped2, and avenue datasets, respectively. Therefore, the proposed model can be employed for accurate and automated anomaly detection in the ITS environment.

Domain adaptation deep learning and its T-S diagnosis networks for the cross-control and cross-condition scenarios in data center HVAC systems

Deep learning has the inspiring potential for artificial intelligence (AI) control in data centers of smart city. However, deep learning-based method is limited when the application scenarios are cross-control or cross-conditions. This paper proposes a novel domain adaptation deep learning model and Teacher-Student (T-S) networks to diagnose the faults of Heating, Ventilation and Air Conditioning (HVAC) systems under cross-control modes and cross-conditions. Firstly, domain adaptation deep belief network (DBN) model is presented to enhance the cross-control generalization performance of extreme learning machine (ELM) and K-nearest neighbor (KNN) diagnosis model. Secondly, the Teacher-Student networks are constructed. With the weights learned from the DBN model in Teacher Network, the cross-conditions capacity is improved for DELM model in the Student Network. Moreover, both the single and multiple faults are experimented and validated, which includes refrigerant leakage, reduced condenser water flow and reduced evaporator flow. The experimental results show that the domain adaptation deep learning model and its T-S diagnosis networks can reach the satisfied generalization performance. For the cross-control tests, the Accuracy of proposed method can reach 91%, 89% and 80% for fault-free, REW and RCW faults. For the cross-temperature and cross-load tests, the proposed method also obtained the high Accuracy of 98% and 88%, respectively.

A hybrid physics-based and data-driven method for gear contact fatigue life prediction

A hybrid physics-based and data-driven method is proposed for gear contact fatigue life prediction. The parameters influencing the fatigue life are determined by the physics-based model. A deep belief network (DBN) model is developed to reveal the relationships between these parameters and fatigue life. A variational autoencoder (VAE) model is presented to expand the size of the training dataset. The proposed method is verified by a gear contact fatigue test, and the predictions are all within a factor of 1.5 scatter band of the experimental results. This work provides an effective method for life prediction with small sample sets.

A novel intelligent approach for predicting meteorological drought based on satellite-based precipitation product: Application of an EMD-DFA-DBN hybrid model

This study uses the long-term Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) dataset with 0.05° spatial resolution to calculate the Standardized Precipitation Index (SPI) signal. The combination of Empirical Mode Decomposition (EMD), Detrended Fluctuation Analysis (DFA) and Deep Belief Network (DBN) as a hybrid model is applied to predict the SPI in the short and long term, i.e. lead times of one, six and twelve months. The performance of the hybrid EMD-DFA-DBN prediction model is evaluated against the hybrid EMD-DFA-MLP model and the stand-alone MLP (Multi-Layer Perceptron) and DBN models. The SPI signal, reconstructed using the EMD-DFA pre-processing method, where the EMD method decomposes the SPI signal into Intrinsic Mode Functions (IMFs) and the DFA method separates high-frequency noisy IMFs from low-frequency useful ones (SNR ≈ 6.49, PSNR ≈ 15.85, CC ≈ 0.88), has lower noise than the original signal; this allows for more accurate predictions of the SPI signal: the DBN-based prediction, with the algorithm trained using the reconstructed SPI signal (CC ≈ 0.94–0.98, RMSE ≈ 0.23–0.38) is superior to that obtained with the algorithm trained using the original SPI signal (CC ≈ 0.87–0.90, RMSE ≈ 0.47–0.51), as well as the MLP-based prediction with training on the reconstructed SPI signal (CC ≈ 0.92–0.95, RMSE ≈ 0.24–0.42) or the original SPI signal (CC ≈ 0.87–0.88, RMSE ≈ 0.48–0.51). It is found that selecting an appropriate stopping criterion for the sifting process is crucial to correctly decompose and reconstruct SPI with the EMD-DFA method. Using the EMD-DFA-DBN hybrid model reduces the noise but preserves useful information in the SPI signal and allows for accurate prediction of drought through deep learning.

Pelican Optimization Algorithm with Federated Learning Driven Attack Detection model in Internet of Things environment

The Internet of Things (IoT) is comprised of millions of physical devices interconnected with the Internet through network that performs a task independently with less human interference. Despite the benefits of IoT, it is vulnerable to malicious attacks. Therefore, machine learning (ML) inspired decision-making is hardly preferred due to drawbacks (the requirement that every training dataset is stored on the central database), computation cost related to the training of massive amounts of information on the unified server, and security issues related to transferring attained statistics from IoT sensor to the server. The Federated Learning (FL) algorithm is the adaptable and most prominent way to resolve the shortcoming of ML-based techniques. The study proposes a Pelican Optimization Algorithm with Federated Learning Driven Attack Detection and classification (POAFL-DDC) technique in the IoT. The POAFL-DDC technique operates on decentralized on-device data for attack detection in the IoT. By substituting update weight with the central FL server, the data is placed on the local IoT device whereas federating training cycles on the DL model. For attack detection process, deep belief network (DBN) model is used in this work. In this study, the POA is utilized to optimize the DBN hyperparameters. The experimental validation of the POAFL-DDC algorithm is tested on TON_IOT dataset and the results are examined in terms of different measures. The experimental outcomes demonstrated that the POAFL-DDC technique reaches superior results over other models.

A hybrid intelligence model for predicting dissolved oxygen in aquaculture water

Dissolved oxygen is an important water quality indicator that affects the health of aquatic products in aquaculture, and its monitoring and prediction are of great significance. To improve the prediction accuracy of dissolved oxygen water quality series, a hybrid prediction model based on variational mode decomposition (VMD) and a deep belief network (DBN) optimized by an improved slime mould algorithm (SMA) is proposed in this paper. First, VMD is used to decompose the nonlinear dissolved oxygen time series into several relatively stable intrinsic mode function (IMF) subsequences with different frequency scales. Then, the SMA is improved by applying elite opposition-based learning and nonlinear convergence factors to increase its population diversity and enhance its local search and global convergence capabilities. Finally, the improved SMA is used to optimize the hyperparameters of the DBN, and the aquaculture water quality prediction VMD-ISMA-DBN model is constructed. The model is used to predict each IMF subsequence, and the ISMA optimization algorithm is used to adaptively select the optimal hyperparameters of the DBN model, and the prediction results of each IMF are accumulated to obtain the final prediction result of the dissolved oxygen time series. The dissolved oxygen data of aquaculture water from 8 marine ranches in Shandong Province, China were used to verify the prediction performance of the model. Compared with the stand-alone DBN model, the prediction performance of the model has been significantly improved, MAE and MSE have been reduced by 43.28% and 40.43% respectively, and (R2) has been increased by 8.37%. The results show that the model has higher prediction accuracy than other commonly used intelligent models (ARIMA, RF, TCN, ELM, GRU and LSTM); hence, it can provide a reference for the accurate prediction and intelligent regulation of aquaculture water quality.

Adaptive Particle Grey Wolf Optimizer with Deep Learning-based Sentiment Analysis on Online Product Reviews

The increasing use of e-commerce websites and social networks is continually generating an immense amount of data in various forms, such as text, images or sounds, videos, etc. Sentiment analysis (SA) in online product reviews is a method of identifying the overall sentiment of customers about a specific product or service. This study used Natural Language Processing (NLP) and Machine Learning (ML) algorithms to identify and extract opinions and emotions expressed in text. Online reviews are often written in informal language, slang, and dialects, making it difficult for ML models to accurately classify sentiments. In addition, the use of misspelled words or incorrect grammar can further complicate the analysis. The recent developments of Deep Learning (DL) models can be used for the accurate classification of sentiments. This paper presents an Adaptive Particle Grey Wolf Optimizer with Deep Learning Based Sentiment Analysis (APGWO-DLSA) method to accurately classify sentiments in product reviews. Initially, data pre-processing was performed to improve the quality of the product reviews using the word2vec embedding process. For sentiment classification, the proposed method used a Deep Belief Network (DBN) model. Finally, the hyperparameter tuning of the DBN was performed using the APGWO algorithm. An extensive experimental analysis demonstrated the improved results of APGWO-DLSA over other methods, showing a maximum accuracy of 94.77% and 85.31% on the Cell Phones And Accessories (CPAA) and Amazon Products (AP) datasets.

Evaluation of Soil Rocky Desertification in Karst Region Based on Deep Belief Network

Abstract Dynamic features from remote sensing photos may be successfully extracted using deep learning and symmetric network structure, which can then be used to direct them to carry out accurate classification. The DBN model can more effectively extract features from photos since it uses unsupervised learning. It can be reduced to the many symmetric Restricted Boltmann Machines (RBM) training problem. In this paper, a soil rocky desertification (RD) assessment model based on a deep belief network (DBN) is created in light of the complicated influencing aspects of Karst RD risk assessment encompassing several geographical elements. The model builds upon the conventional RBM framework and incorporates the influence layer of related elements as an auxiliary requirement for retrieving Geographic Information System (GIS) score data. Then, in order to forecast the level of soil rocky desertification, it learns the features of many elements. The experimental results show that the proposed model proposed in this paper has better prediction performance and faster convergence speed, and its classification results for different degrees of RD are more consistent with the actual risk assessment results.