Deep Convolutional Neural Network Algorithm Research Articles

A trusted distributed routing scheme for wireless sensor networks using blockchain and meta‐heuristics‐based deep learning technique

AbstractThe wireless sensor network (WSN) with fluctuating environs might be susceptible to diverse types of malicious cyber‐attacks, and they are mostly dependent on the authentication and encryption algorithm to astound this challenge. Most predominant routing schemes in literature are fall backs in characterizing the malicious nodes on networks due to the real time variation of routing information. Therefore, a reliable and trustworthy inter‐correlated routing scheme based on Block chain, Meta‐heuristic, and Deep Learning Algorithms are presented in this paper. The disseminated routing info in the WSN is handled by Block chain strategy, in which the optimal routing is accomplished with the help of Salp Swarm Optimization algorithm. The routing info variations between the nodes are envisaged and the optimal routing decisions are done by using the Deep Convolutional Neural network algorithm. The proposed routing scheme is implemented in NS2 and its performance is evaluated based on latency, energy consumption, and throughput metrics are analyzed. The efficiency of the method is improved as 97% and the evaluation is done for the malicious attacks, latency, and the delay. The comparison is made for the existing methods as particle swarm optimization, Markov decision process, security disjoint routing‐based verified message, trusted‐cluster–based routing, and reinforcement learning‐based neural network (RLNN) with the proposed method for the delay ratio.

Validation and estimation of spleen volume via computer-assisted segmentation on clinically acquired CT scans.

Purpose: Deep learning is a promising technique for spleen segmentation. Our study aims to validate the reproducibility of deep learning-based spleen volume estimation by performing spleen segmentation on clinically acquired computed tomography (CT) scans from patients with myeloproliferative neoplasms. Approach: As approved by the institutional review board, we obtained 138 de-identified abdominal CT scans. A sum of voxel volume on an expert annotator's segmentations establishes the ground truth (estimation 1). We used our deep convolutional neural network (estimation 2) alongside traditional linear estimations (estimation 3 and 4) to estimate spleen volumes independently. Dice coefficient, Hausdorff distance, coefficient, Pearson coefficient, the absolute difference in volume, and the relative difference in volume were calculated for 2 to 4 against the ground truth to compare and assess methods' performances. We re-labeled on scan-rescan on a subset of 40 studies to evaluate method reproducibility. Results: Calculated against the ground truth, the coefficients for our method (estimation 2) and linear method (estimation 3 and 4) are 0.998, 0.954, and 0.973, respectively. The Pearson coefficients for the estimations against the ground truth are 0.999, 0.963, and 0.978, respectively (paired -tests produced between 2 and 3, and 2 and 4). Conclusion: The deep convolutional neural network algorithm shows excellent potential in rendering more precise spleen volume estimations. Our computer-aided segmentation exhibits reasonable improvements in splenic volume estimation accuracy.

Deep Learning Algorithm Trained with COVID-19 Pneumonia Also Identifies Immune Checkpoint Inhibitor Therapy-Related Pneumonitis.

Simple SummaryThe use of immune checkpoint inhibitors (ICIs) to treat oncologic diseases is progressively increasing. Computed tomography (CT) features of ICI therapy-related pneumonitis may overlap with other diseases, including coronavirus disease 2019 (COVID-19). Thus, oncologic patients undergoing ICI therapy and developing pneumonitis are at risk of being misdiagnosed. Exploring the strengths and weaknesses of artificial intelligence in distinguishing between ICI therapy-related pneumonitis and COVID-19 is of great importance for oncologic patients and for clinicians in order to increase awareness on this topic and stimulate novel strategies aimed to promptly and correctly classify and treat this category of vulnerable patients.Background: Coronavirus disease 2019 (COVID-19) pneumonia and immune checkpoint inhibitor (ICI) therapy-related pneumonitis share common features. The aim of this study was to determine on chest computed tomography (CT) images whether a deep convolutional neural network algorithm is able to solve the challenge of differential diagnosis between COVID-19 pneumonia and ICI therapy-related pneumonitis. Methods: We enrolled three groups: a pneumonia-free group (n = 30), a COVID-19 group (n = 34), and a group of patients with ICI therapy-related pneumonitis (n = 21). Computed tomography images were analyzed with an artificial intelligence (AI) algorithm based on a deep convolutional neural network structure. Statistical analysis included the Mann–Whitney U test (significance threshold at p < 0.05) and the receiver operating characteristic curve (ROC curve). Results: The algorithm showed low specificity in distinguishing COVID-19 from ICI therapy-related pneumonitis (sensitivity 97.1%, specificity 14.3%, area under the curve (AUC) = 0.62). ICI therapy-related pneumonitis was identified by the AI when compared to pneumonia-free controls (sensitivity = 85.7%, specificity 100%, AUC = 0.97). Conclusions: The deep learning algorithm is not able to distinguish between COVID-19 pneumonia and ICI therapy-related pneumonitis. Awareness must be increased among clinicians about imaging similarities between COVID-19 and ICI therapy-related pneumonitis. ICI therapy-related pneumonitis can be applied as a challenge population for cross-validation to test the robustness of AI models used to analyze interstitial pneumonias of variable etiology.

WITHDRAWN: Experimental analysis of brain tumor detection system using Machine learning approach

In the bio-medical digital image processing industry, tumors are the major cause to deal with and analysis of tumors is the crucial part as well. Many critical system learning approaches are designed over past several years to identify the prediction accuracy improvement ratio of such tumors, still it is a complex task to finalize. So, that a new machine learning approach is required to resolve the issue of identifying the tumors in earlier stages. The term Gliomas creates a major alert in medical industry, which is one of the powerful and common brain tumor category. In general Magnetic-Resonance-Imaging (MRI) is a generally known image processing strategy to evaluate such kinds of tumors, however the huge measure of information delivered by MRI forestalls 'manual segmentation' problems in a sensible time, restricting the utilization of exact quantitative estimations in the medical field. In this paper, a new systematic and strategy is proposed based on Deep Convolutional Neural Network Algorithm (DCNNA), which explores all kinds of 3X3 kernel portions. The proposed approach of Brain Tumor Segmentation using DCNNA validates the process based on BRATS dataset and attaining the good result based on similarity co-efficient metrics over the ranges of 0.886, 0.833 and 0.773. As well as, the same metrics are sustained in the first category of the Web based Tumor Investigation environment. A fuzzy based approach is also added into the proposed processing stages of segmenting on brain tumor classifications, which will improve the accuracy ratio of the proposed DCNNA approach as well as the Non Linear Approach shows the classical deviations over results and the proposed algorithm part of this paper clearly summarizes the features well. The resulting accuracy of the proposed brain tumor segmentation is also comparatively good in state, which will be showed on resulting section of this paper.

Intelligent Mapping of Urban Forests from High-Resolution Remotely Sensed Imagery Using Object-Based U-Net-DenseNet-Coupled Network

The application of deep learning techniques, especially deep convolutional neural networks (DCNNs), in the intelligent mapping of very high spatial resolution (VHSR) remote sensing images has drawn much attention in the remote sensing community. However, the fragmented distribution of urban land use types and the complex structure of urban forests bring about a variety of challenges for urban land use mapping and the extraction of urban forests. Based on the DCNN algorithm, this study proposes a novel object-based U-net-DenseNet-coupled network (OUDN) method to realize urban land use mapping and the accurate extraction of urban forests. The proposed OUDN has three parts: the first part involves the coupling of the improved U-net and DenseNet architectures; then, the network is trained according to the labeled data sets, and the land use information in the study area is classified; the final part fuses the object boundary information obtained by object-based multiresolution segmentation into the classification layer, and a voting method is applied to optimize the classification results. The results show that (1) the classification results of the OUDN algorithm are better than those of U-net and DenseNet, and the average classification accuracy is 92.9%, an increase in approximately 3%; (2) for the U-net-DenseNet-coupled network (UDN) and OUDN, the urban forest extraction accuracies are higher than those of U-net and DenseNet, and the OUDN effectively alleviates the classification error caused by the fragmentation of urban distribution by combining object-based multiresolution segmentation features, making the overall accuracy (OA) of urban land use classification and the extraction accuracy of urban forests superior to those of the UDN algorithm; (3) based on the Spe-Texture (the spectral features combined with the texture features), the OA of the OUDN in the extraction of urban land use categories can reach 93.8%, thereby the algorithm achieved the accurate discrimination of different land use types, especially urban forests (99.7%). Therefore, this study provides a reference for feature setting for the mapping of urban land use information from VHSR imagery.

CNN based approach for identifying banana species from fruits

Classification and recognition of fruits are still the challenging one as there are different classes of fruit types having wide inter-class resemblance. This paper proposes a banana identification model using a five-layer convolution neural network (CNN) which composed of convolution layer, pooling layer and fully connected layer. Fruits like Apple, Strawberry, and Orange, Mango and banana have been analyzed and several features have been extracted using deep learning-based CNN algorithm. Finally, the fruit identification process is done by random forest and K-Nearest Neighborhood (K-NN) classifying algorithms. A regular digital camera is used for the image acquisition process. All image manipulation processes are performed in MATLAB-17 environment.Experiments are conducted in our database consisting of 5887 fruit images. The performance of the proposed deep learning based-random forest and KNN classifiers are compared with the existing feature extraction method of HOG based random forest and KNN in terms of accuracy, precision, recall and f-score and we have achieved a accuracy rate of 96.98% for deep feature-random forest classification combination algorithm which is better than the deep feature-KNN, HOG based KNN and random forest classifiers.

Landmark annotation and mandibular lateral deviation analysis of posteroanterior cephalograms using a convolutional neural network

Background/purposeFacial asymmetry is relatively common in the general population. Here, we propose a fully automated annotation system that supports analysis of mandibular deviation and detection of facial asymmetry in posteroanterior (PA) cephalograms by means of a deep learning-based convolutional neural network (CNN) algorithm. Materials and methodsIn this retrospective study, 400 PA cephalograms were collected from the medical records of patients aged 4 years 2 months–80 years 3 months (mean age, 17 years 10 months; 255 female patients and 145 male patients). A deep CNN with two optimizers and a random forest algorithm were trained using 320 PA cephalograms; in these images, four PA landmarks were independently identified and manually annotated by two orthodontists. ResultsThe CNN algorithms had a high coefficient of determination (R2), compared with the random forest algorithm (CNN-stochastic gradient descent, R2 = 0.715; CNN-Adam, R2 = 0.700; random forest, R2 = 0.486). Analysis of the best and worst performances of the algorithms for each landmark demonstrated that the right latero-orbital landmark was most difficult to detect accurately by using the CNN. Based on the annotated landmarks, reference lines were defined using an algorithm coded in Python. The CNN and random forest algorithms exhibited similar accuracy for the distance between the menton and vertical reference line. ConclusionOur findings imply that the proposed deep CNN algorithm for detection of facial asymmetry may enable prompt assessment and reduce the effort involved in orthodontic diagnosis.

Improvement of the AlexNet Networks for Large-Scale Recognition Applications

In this paper, the deep convolutional neural networks (DCNNs) are studied to perform the complex feature extraction on the image in the convolution layer and to improve the final test accuracy of the network. By improving the DCNNs algorithm and framework, it can enhance the accurate extraction of the image features. We replace the fully connection layer of the original network with the global average pooling layer. In the absence of the large number of calculations of network parameters, the final effect is not changed; thereby, it increases the speed of the network. The simulation result is given to show the effectiveness of the DCNNs algorithm by comparing the training accuracy and test accuracy of the five improvement algorithms.

Development of an Artificial Intelligence Model to Identify a Dental Implant from a Radiograph.

The objective of this study was to develop a deep convolutional neural network (CNN) that would identify the brand and model of a dental implant from a radiograph. A data augmentation procedure provided a total of 1,206 dental implant radiographic images of three different brands for six models (Nobel Biocare NobelActive [NNA] and Br.nemark System [NBS], Straumann Bone Level [SBL] and Tissue Level [STL], and Zimmer Biomet Dental Tapered Screw-Vent [ZTSV] and SwissPlus [ZSP]). They were divided into a test group (n = 241; 19.9%) and a training and validation group (n = 965; 80%). Preprocessing and transfer learning were applied to a pretrained GoogLeNet Inception CNN network. The accuracy, sensitivity, specificity, positive predictive value, negative predictive value, receiver operating characteristic (ROC) curve, and area under the ROC curve (AUC) of the CNN model were evaluated. The diagnostic accuracy was 93.8% (95% CI: 87.2% to 99.4%), the sensitivity was 93.5% (95% CI: 84.2% to 99.3%), the specificity was 94.2% (95% CI: 83.5% to 99.4%), the positive predictive value was 92% (95% CI: 83.9% to 97.2%), and the negative predictive value was 91.5% (95% CI: 80.2% to 97.1%). The deep CNN algorithm achieved an AUC of 0.918 (95% CI: 0.826 to 0.973) on NNA, 0.922 (95% CI: 0.831 to 0.964) on NBS, 0.909 (95% CI: 0.844 to 0.982) on SBL, 0.890 (95% CI: 0.783 to 0.945) on STL, 0.931 (95% CI: 0.867 to 0.979) on ZTSV, and 0.911 (95% CI: 0.811 to 0.957) on ZSP. The deep CNN model had a very good performance in identifying a dental implant from a radiograph. A huge and varied database of radiographs would have to be built up to be able to identify any dental implant.

Energy-Efficient Wearable EPTS Device Using On-Device DCNN Processing for Football Activity Classification

This paper presents an energy-optimized electronic performance tracking system (EPTS) device for analyzing the athletic movements of football players. We first develop a tiny battery-operated wearable device that can be attached to the backside of field players. In order to analyze the strategic performance, the proposed wearable EPTS device utilizes the GNSS-based positioning solution, the IMU-based movement sensing system, and the real-time data acquisition protocol. As the life-time of the EPTS device is in general limited due to the energy-hungry GNSS sensing operations, for the energy-efficient solution extending the operating time, in this work, we newly develop the advanced optimization methods that can reduce the number of GNSS accesses without degrading the data quality. The proposed method basically identifies football activities during the match time, and the sampling rate of the GNSS module is dynamically relaxed when the player performs static movements. A novel deep convolution neural network (DCNN) is newly developed to provide the accurate classification of human activities, and various compression techniques are applied to reduce the model size of the DCNN algorithm, allowing the on-device DCNN processing even at the memory-limited EPTS device. Experimental results show that the proposed DCNN-assisted sensing control can reduce the active power by 28%, consequently extending the life-time of the EPTS device more than 1.3 times.

Deep learning method for segmentation of rotator cuff muscles on MR images.

To develop and validate a deep convolutional neural network (CNN) method capable of (1) selecting a specific shoulder sagittal MR image (Y-view) and (2) automatically segmenting rotator cuff (RC) muscles on a Y-view. We hypothesized a CNN approach can accurately perform both tasks compared with manual reference standards. We created 2 models: model A for Y-view selection and model B for muscle segmentation. For model A, we manually selected shoulder sagittal T1 Y-views from 258 cases as ground truth to train a classification CNN (Keras/Tensorflow, Inception v3, 16 batch, 100 epochs, dropout 0.2, learning rate 0.001, RMSprop). A top-3 success rate evaluated model A on 100 internal and 50 external test cases. For model B, we manually segmented subscapularis, supraspinatus, and infraspinatus/teres minor on 1048 sagittal T1 Y-views. After histogram equalization and data augmentation, the model was trained from scratch (U-Net, 8 batch, 50 epochs, dropout 0.25, learning rate 0.0001, softmax). Dice (F1) score determined segmentation accuracy on 105 internal and 50 external test images. Model A showed top-3 accuracy > 98% to select an appropriate Y-view. Model B produced accurate RC muscle segmentations with mean Dice scores > 0.93. Individual muscle Dice scores on internal/external datasets were as follows: subscapularis 0.96/0.93, supraspinatus 0.97/0.96, and infraspinatus/teres minor 0.97/0.95. Our results show overall accurate Y-view selection and automated RC muscle segmentation using a combination of deep CNN algorithms.

Comparisons of Facial Recognition Algorithms Through a Case Study Application

&lt;p class="affiliations"&gt;&lt;strong&gt;Abstract— &lt;/strong&gt;Computer visions and their applications have become important in contemporary life. Hence, researches on facial and object recognition have become increasingly important both from academicians and practitioners. Smart gadgets such as smartphones are nowadays capable of high processing power, memory capacity, along with high resolutions camera. Furthermore, the connectivity bandwidth and the speed of the interaction have significantly impacted the popularity of mobile object recognition applications. These developments in addition to computer vision’s algorithms advancement have transferred object’s recognitions from desktop environments to the mobile world. The aim of this paper to reveal the efficiency and accuracy of the existing open-source facial recognition algorithms in real-life settings. We use the following popular open-source algorithms for efficiency evaluations: Eigenfaces, Fisherfaces, Local Binary Pattern Histogram, the deep convolutional neural network algorithm, and OpenFace. The evaluations of the test cases indicate that among the compared facial recognition algorithms the OpenFace algorithm has the highest accuracy to identify faces. The findings of this study help the practitioner on their decision of the algorithm selections and the academician on how to improve the accuracy of the current algorithms even further.&lt;/p&gt;

Efficacy of deep convolutional neural network algorithm for the identification and classification of dental implant systems, using panoramic and periapical radiographs: A pilot study.

Convolutional neural networks (CNNs), a particular type of deep learning architecture, are positioned to become one of the most transformative technologies for medical applications. The aim of the current study was to evaluate the efficacy of deep CNN algorithm for the identification and classification of dental implant systems.A total of 5390 panoramic and 5380 periapical radiographic images from 3 types of dental implant systems, with similar shape and internal conical connection, were randomly divided into training and validation dataset (80%) and a test dataset (20%). We performed image preprocessing and transfer learning techniques, based on fine-tuned and pre-trained deep CNN architecture (GoogLeNet Inception-v3). The test dataset was used to assess the accuracy, sensitivity, specificity, receiver operating characteristic curve, area under the receiver operating characteristic curve (AUC), and confusion matrix compared between deep CNN and periodontal specialist.We found that the deep CNN architecture (AUC = 0.971, 95% confidence interval 0.963–0.978) and board-certified periodontist (AUC = 0.925, 95% confidence interval 0.913–0.935) showed reliable classification accuracies.This study demonstrated that deep CNN architecture is useful for the identification and classification of dental implant systems using panoramic and periapical radiographic images.

Uncertainty quantification in deep convolutional neural network diagnostics of journal bearings with ovalization fault

Bearings play a crucial role in machine longevity and is, at the same time, one of the most critical sources of failure in rotor dynamics. Particularly for journal bearings, it is not completely understood how specific damages may influence the response of the rotating system. Consequently, the identification of hydrodynamic bearing faults is challenging. Most of the literature relies on large amounts of training data collections from physical experiments or from the field, which are high in cost. This paper offers a deep learning approach to identify ovalization faults aiming to develop condition monitoring model-based strategies applied to hydrodynamic journal bearings. Therefore, a numerical model was developed to simulate the ovalization fault conditions in order to build training datasets. Afterwards, a deep convolutional neural network algorithm was trained with the generated datasets and used to predict the faults conditions. Finally, the identification performance was evaluated statistically regarding the true-positive identification by both probability density function and subjective logic. The classification accuracy showed promising results for training the machine learning algorithms with simulated data.

Deep locality‐sensitive discriminative dictionary learning for semantic video analysis

SummaryVideo semantic analysis (VSA) has received significant attention in the area of Machine Learning for some time now, particularly video surveillance applications with sparse representation and dictionary learning. Studies have shown that the duo has significantly impacted on the classification performance of video detection analysis. In VSA, the locality structure of video semantic data containing more discriminative information is very essential for classification. However, there has been modest feat by the current SR‐based approaches to fully utilize the discriminative information for high performance. Furthermore, similar coding outcomes are missing from current video features with the same video category. To handle these issues, we first propose an improved deep learning algorithm—locality deep convolutional neural network algorithm (LDCNN) to better extract salient features and obtain local information from semantic video. Second, we propose a novel DL method, called deep locality‐sensitive discriminative dictionary learning (DLSDDL) for VSA. In the proposed DLSDDL, a discriminant loss function for the video category based on sparse coding of sparse coefficients is introduced into the structure of the locality‐sensitive dictionary learning (LSDL) method. After solving the optimized dictionary, the sparse coefficients for the testing video feature samples are obtained, and then the classification result for video semantic is realized by reducing the error existing between the original and recreated samples. The experiment results show that the proposed DLSDDL technique considerably increases the efficiency of video semantic detection as against competing methods used in our experiment.

Fast location and classification of small targets using region segmentation and a convolutional neural network

The hickory nut has rich nutritional and high economic value, but its inner wall structure is complex, making it inconvenient to separate the kernel from the hickory nut fragments for consumption. In this study, we apply a deep convolutional neural network algorithm to hickory nut fragments after shell breaking, to accurately sort kernels, shells, and unseparated bodies. Region segmentation is used to eliminate interference when locating multiple small targets of different sizes, and a custom dataset of 15,000 target images after shell breaking is created. Then, three different loss function optimization methods are evaluated for model optimization. Results revealed that the model only minimizes the average cross-entropy loss, and that using dropout at the first fully connected layer with a dropout rate 0.1 provides the best shell, kernel, and unseparated body identification accuracy of 97.14%, 99.66%, and 96.48%, respectively. Notably, the kernel and unseparated body recognition rates are considerably improved over other methods. Finally, a total of 450 hickory nut fragments were divided into 30 batches for a sorting experiment. The resulting shell, kernel, and unseparated body identification accuracy is 94.00%, 98.67%, and 95.33%, respectively. These results demonstrate that the proposed method provides excellent classification of hickory nut fragments according to the material. This approach can be easily applied to any small target sorting application.

Self-Learning Perfect Optical Chirality via a Deep Neural Network.

Optical chirality occurs when materials interact differently with light in a specific circular polarization state. Chiroptical phenomena inspire wide interdisciplinary investigations, which require advanced designs to reach strong chirality for practical applications. The development of artificial intelligence provides a new vision for the manipulation of light-matter interaction beyond the theoretical interpretation. Here, we report a self-consistent framework named the Bayesian optimization and convolutional neural network that combines Bayesian optimization and deep convolutional neural network algorithms to calculate and optimize optical properties of metallic nanostructures. Both electric-field distributions at the near field and reflection spectra at the far field are calculated and self-learned to suggest better structure designs and provide possible explanations for the origin of the optimized properties, which enables wide applications for future nanostructure analysis and design.

Effective textile quality processing and an accurate inspection system using the advanced deep learning technique

This research paper focuses on the innovative detection of defects in fabric. This approach is based on the design and development of a computer-assisted system using the deep learning technique. The classification network is modeled using the ResNet512-based Convolutional Neural Network to learn the deep features in the presented fabric. Being an accurate method, this enables accurate localization of minute defects too. Our classification is based on three major steps; firstly, an image acquired by the NI Vision model and pre-processed for a standard pattern to Kullback Leibler Divergence calculation. Secondly, standard textile fabrics are presented to train the Convolutional Neural Network to classify the defective region and the defect-free region. Finally, the testing fabrics are examined by the trained deep Convolutional Neural Network algorithm. To verify the performance, multiple fabrics are presented and the classification accuracy is evaluated. For standard defects on defective fabrics, an average accuracy of 96.5% with 98.5% precision is obtained. Experimental results on the standard Textile Texture Database dataset confirmed that our method provides better results compared with similar recent classification methods, such as the Support Vector Machine and Bayesian classifier.

OPU: An FPGA-Based Overlay Processor for Convolutional Neural Networks

Field-programmable gate array (FPGA) provides rich parallel computing resources with high energy efficiency, making it ideal for deep convolutional neural network (CNN) acceleration. In recent years, automatic compilers have been developed to generate network-specific FPGA accelerators. However, with more cascading deep CNN algorithms adapted by various complicated tasks, reconfiguration of FPGA devices during runtime becomes unavoidable when network-specific accelerators are employed. Such reconfiguration can be difficult for edge devices. Moreover, network-specific accelerator means regeneration of RTL code and physical implementation whenever the network is updated. This is not easy for CNN end users. In this article, we propose a domain-specific FPGA overlay processor, named OPU to accelerate CNN networks. It offers software-like programmability for CNN end users, as CNN algorithms are automatically compiled into executable codes, which are loaded and executed by OPU without reconfiguration of FPGA for switch or update of CNN networks. Our OPU instructions have complicated functions with variable runtimes but a uniform length. The granularity of instruction is optimized to provide good performance and sufficient flexibility, while reducing complexity to develop microarchitecture and compiler. Experiments show that OPU can achieve an average of 91% runtime multiplication and accumulation unit (MAC) efficiency (RME) among nine different networks. Moreover, for VGG and YOLO networks, OPU outperforms automatically compiled network-specific accelerators in the literature. In addition, OPU shows $5.35\times $ better power efficiency compared with Titan Xp. For a real-time cascaded CNN networks scenario, OPU is $2.9\times $ faster compared with edge computing GPU Jetson Tx2, which has a similar amount of computing resources.

Applications of deep convolutional neural networks in prospecting prediction based on two-dimensional geological big data

There are many challenges in the task of predicting ore deposits from big data repositories. The data are inherently complex and of great significance to the intervenient spatial relevance of deposits. The characteristics of the data make it difficult to use machine learning algorithms for the quantitative prediction of mineral resources. There are considerable interest and value in extracting spatial distribution characteristics from two-dimensional (2-d) ore-controlling factor layers under different metallogenic conditions. In this paper we undertake such analysis using a deep convolutional neural network algorithm named AlexNet. Training on the 2-d mineral prediction and classification model is performed using data from the Songtao–Huayuan sedimentary manganese deposit. It mines the coupling correlation between the spatial distribution of chemical elements, sedimentary facies, the outcrop of Datangpo Formation, faults, water system, and the areas where manganese ore bodies are present, as well as the correlation among different ore-controlling factors by employing the AlexNet networks. By comparing the training loss, training accuracy, verification accuracy, and recall of models trained by different scales of grids and different combinations of ore-controlling factor layers, we further discuss the most appropriate scale division and the optimal combination of ore-controlling factors to make the model achieve its strongest robustness. It is found that the prediction performance of AlexNet networks reaches its peak when selecting a grid division of 200 pixels × 200 pixels (the actual distance is 10 km × 10 km) and inputting the distribution layers of 21 chemical elements maps, lithofacies–paleogeographic map, formation and tectonic map, outcrop map of Datangpo Formation, and water system map. The training loss, training accuracy, verification accuracy, and recall of the optimal model are 0.0000001, 100.00%, 86.21%, and 91.67%, respectively. The proposed method is successfully applied to the 2-d metallogenic prediction in Songtao–Huayuan study area. And five metallogenic prospective areas from A to E are delineated with large probability for potential ore bodies.