Search

The increasing availability of hydrological and physiographic spatiotemporal data has boosted machine learning's role in rapid flood mapping. Yet, data scarcity, especially high-resolution DEMs, challenges regions with limited access. This paper examines how DEM type and resolution affect flood prediction accuracy, utilizing a cutting-edge deep learning (DL) method called 1D convolutional neural network (CNN). It utilizes synthetic hydrographs as training input and water depth data obtained from LISFLOOD-FP, a 2D hydrodynamic model, as target data. This study investigates digital surface models (DSMs) and digital terrain models (DTMs) derived from a 1 m LIDAR-based DTM, with resolutions from 15 to 30 m. The methodology is applied and assessed in a established benchmark, city of Carlisle, UK. The models' performance is then evaluated and compared against an observed flood event using RMSE, Bias, and Fit indices. Leveraging the insights gained from this region, the paper discusses the applicability of the methodology to address the challenges encountered in a data-scarce flood-prone region, exemplified by Pakistan. Results indicated that utilizing a 30 m DTM outperformed a 30 m DSM in terms of flood depth prediction accuracy by about 21% during the flood peak stage, highlighting the superior performance of DTM at lower resolutions. Increasing the resolution of DTM to 15 m resulted in a minimum 50% increase in RMSE and a 20% increase in fit index across all flood stages. The findings emphasize that while a coarser resolution DEM may impact the accuracy of machine learning models, it remains a viable option for rapid flood prediction. However, even a slight improvement in data resolution in data-scarce regions would provide significant added value, ultimately enhancing flood risk management. [ABSTRACT FROM AUTHOR]

The world has moved into an era of hidden ore body exploration, necessitating the development of new prospecting and exploration methods. One promising approach is to use the deep convolutional neural network (DCNN) algorithm to extract spatial and correlation characteristics of multiple two-dimensional elements related to hidden ores. This paper explores this method on Datangpo manganese (Mn), constructing prediction datasets that includes geological, geochemical, geophysical and aeromagnetic features. Analyzing metallogenic conditions and control factors of Mn ores, we construct a Mn ore prediction model (Geo-DCNN) based on multiple geographical knowledge and DCNN. The Geo-DCNN model reaches ore-bearing accuracy of 79.11%, non-ore-bearing accuracy of 99.01%, overall accuracy of 95.35%, and loss value of 0.0227 after training. Based on analysis of ROC curve, P-R curve, field investigation, and target area verification, we discover that the prediction results of the Geo-DCNN model in northeast Guizhou have a high correspondence rate with known manganese deposits. This provides valuable insight for further ore exploration in the area. Additionally, the results indicate that the Geo-DCNN model is robust and portable, suggesting that it can be applied to metallogenic prediction practices for manganese ore in similar regions. [ABSTRACT FROM AUTHOR]

Conventional convolutional neural networks (CNN) are deficient in the rock type recognition due to large convolutional kernels and numerous network parameters necessitated for recognition of complex images. The advanced convolutional neural network, Visual Geometry Group-16 (VGG16) model, which is based on multiple small convolutional kernels and fully connected layers, attains higher classification accuracy, yet is limited by low computational efficiency. In this paper, we propose a novel approach that integrates the advanced VGG16 with the Principal Component Analysis (PCA), a dimensionality reduction technique. This integration, referred to as the PCA-VGG16 model, aims to enhance the computational efficiency of automatic rock type identification. A dataset comprising 3000 images of six rock types: limestone, shale, dolomite, quartzite, marble, and granite, is assembled for training and testing of the PCA-VGG16 model. The feasibility of the PCA-VGG16 model for classification prediction is demonstrated through evaluation metrics including accuracy, loss value, and F1-score. A comparative analysis with the CNN and VGG16 models reveals that the proposed PCA-VGG16 model exhibits superior classification accuracy and reduced training durations, making a very promising advancement in the field. Furthermore, an in-depth analysis is conducted to understand the impact of dataset size and key hyperparameters (such as epochs and batch size) on the classification accuracy of the PCA-VGG16 model. The findings indicate that a minimum dataset of 1500 sample images is necessary to achieve a classification accuracy above 90%. For optimal model performance, a division of training, validation, and test sets approximately at 6:2:2, along with two epochs and a batch size of 128, is recommended in this study. [ABSTRACT FROM AUTHOR]

It is of great significance to accurately grasp ground-level ozone concentration on a wide scale in order to cope with the increasingly serious ground-level ozone pollution problem. Currently, satellite technology can be used to monitor ground-level ozone concentration on a wide scale, but there is the problem of insufficient spatial resolution. The use of ground monitoring stations to detect ground-level ozone concentration can obtain more accurate ozone concentration around the station, but there are problems of limited number of stations and sparse distribution. Therefore, it is still a challenge to accurately grasp the ground-level ozone concentration at a large scale and high spatial resolution. To address this issue, this paper proposes a ground-level ozone concentration inversion algorithm based on deep learning methods, incorporating data from remote sensing satellites, ground monitoring stations, meteorological conditions, geography, and land use. Taking Yunnan Province in China as the experimental region and utilizing five-fold cross-validation, the model's Mean Square Error (MSE) reached 340.45 and Root Mean Square Error (RMSE) reached 18.45. Compared to machine learning models, the error decreased by 14.53%, outperforming Random Forest models, Convolution Neural Networks, and Vision Transformers. This study offers a referential approach for accurately capturing ground-level ozone concentrations at a large scale with high spatial resolution. [ABSTRACT FROM AUTHOR]

With the increasing proportion of solar energy in the energy system, accurate solar irradiance forecast is of great significance for low-cost energy scheduling. This paper proposes a new forecasting idea for day-ahead solar irradiance forecast on the day-ahead scale: Firstly, based on formula derivation and big data correlation analysis, this paper finds out multiple parameters related to GHI, and jointly uses these parameters to forecast GHI. Error revision during morning period (ERDMP) is innovatively proposed on this basis, towards a more accurate daytime forecast. In order to prove the reliability and universality of the method, relevant data from five different-climatic regions are respectively used in the experiment. The multi-variable historical data-based day-ahead solar irradiance forecast uses deep neural networks, including Recurrent Neural Network (RNN), Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) network. ERDMP uses a linear AutoRegressive (AR) model to predict the daytime error coefficients based on the morning error coefficients. According to the results, through the proposed ERDMP, Mean Absolute Error (MAE) decreases by about 25% to 30%, Root Mean Squared Error (RMSE) decreases by about 20%, and R2 increases by about 5% to 10% when compared with the initial error of multi-parameter prediction models and other advanced models. [ABSTRACT FROM AUTHOR]

In the geological field, recognizing rock thin section images under a microscope is of great significance to geological research and mineral resources exploration. Compared with other images, the rock thin section microscope image has complex features and rich information. When classifying for them, the redundant information and features that are useless for classification tasks will affect the model classification effect. Thus this paper proposes a rock image classification algorithm based on depth residuals shrinkage network and attention mechanism to suppress the useless information. The subnetwork of obtaining threshold value is improved, and the global maximum pooling of features is added as information representation, and the soft threshold function is improved by adding the weight coefficient which is based on attention mechanism to distinguish the importance of different features. Moreover, three rock thin section microscopic image classification algorithms fusing multidimensional information are designed, and the orthogonal polarized image and single polarized image of rock thin section are input as the base data, which make full use of the multidimensional information of rock thin section microscopic image. This study used 20,242 images of 12 kinds of rock thin sections as samples to train and verify the above method. The results show that the method proposed in this paper can effectively improve the recognition accuracy of rock thin section images under a microscope. [ABSTRACT FROM AUTHOR]

The WHO (World Health Organization) study reports that, between 1998-2017, 4.8 million people have been affected by landslides with more than 18000 deaths. The prevailing climate change and erratic high intensity rainfall are expected to trigger more landslides, which can increase the death rates per year in future. Therefore, evolving successful mechanisms to identify and predict landslides is critical in risk reduction and post-disaster management activities. With the applications of Machine Learning , the success rates of landslide identification have been improved significantly. This review paper presents the results of data analysis on the papers published for the last three decades on varying degrees of reliability and success rate on the theme "Machine Learning for landslide identification, mitigation, and prediction". The analyses show how the reliability and accuracy of the landslide prediction model have improved considerably with the tools available in Machine Learning. Though many conventional tools such as statistical packages are available, the Machine Learning algorithms gave a robust dimension for a reliable landslide risk analysis, modeling prediction tools and post-disaster damage identification, This paper recommends a multi-modal framework for characterising landslides in all aspects using Machine Learning techniques that could outperform single modal approaches. Open research problems and future research dimensions by integrating landsat data and Machine Learning for landslide studies are also discussed in this work which would be beneficial for researchers in this field and also to the community at large across the globe. [ABSTRACT FROM AUTHOR]

Crop classification is of great significance as it opens doors to agricultural inventory and research related to crop mapping, crop yield, and economic analysis. In most studies to date, openly accessible satellites like Landsat, Sentinel, and MODIS are generally used. These satellites suffer from low spatial resolution as well as high revisit rates. Further, the classification in these studies is carried out at pixel level, which is not reliable as per-pixel spectral analysis often gives ambiguous results. Thus, this study proposes the use of high-resolution temporal PlanetScope imagery with an average spatial resolution of 3.7 m for object-based image analysis (OBIA) at the Southern Indian basin at the River Krishna. This area is abundant with a variety of crops like paddy, chilli, maize, lily, colocasia, curry leaves, mint, bhindi, banana, betel leaves, and sugarcane. The latest available ground truth data was from 2017 and hence, this has been used. Despite this data being a year older than the one used in a previous study, good and commendable accuracies were still produced by this work. The machine learning (ML) modelling and evaluation was carried out in Python while Google Earth Engine (GEE) was used as the primary platform for feature extraction, dataset preparation, and visualization of the final crop-classified image. For the purpose of object-based classification, this paper puts forward Support Vector Machines (SVM). To provide a comparative analysis to justify the performance of modelled SVM, several other ML algorithms like Convolution Neural Networks (CNN), Random Forest (RF), Artificial Neural Network (ANN), and Bayes Classifier were then modelled for this methodology in this study area. The results show that SVM performed best, with a high accuracy of 94.3%. All other algorithms modelled showed less accuracy comparatively, but were still above 75%. [ABSTRACT FROM AUTHOR]

Building footprints are the most visible features in urban areas. Detecting building footprint has a substantial position in decision-making problems such as city planning and development, urban mapping and management, population estimation, etc. In this paper, we aim to automatically detect and count building footprints by leveraging deep learning techniques and the potential availability of remote sensing datasets at high spatial resolutions. We build object detection deep learning models by integrating the partitioning segmentation and convolutional neural networks (CNN) architecture using the WorldView-3 very high-resolution optical satellite imagery data. Our detection model is implemented on data from the Paris, France, and Khartoum, Sudan regions. These regions were chosen because it has very different characteristics and large data variation, making it challenging to extract building features from satellite imagery. This challenge is even greater because one area is taken perpendicularly and the other area is taken at an angle of several degrees which causes more noise, one of which is the shadow of the building. The detection models can detect and count the building footprints and have promising performance on large variable data with an Average Precision (AP) of 64.19%. Paris was detected better by 74.66% compared to Khartoum by 56.19%. The partitioning segmentation technique is used to tune the anchor boxes of CNN input images. Our result offers a potential practical implementation for rapid yet accurate estimation of urban monitoring and city planning, particularly in metropolitan regions. [ABSTRACT FROM AUTHOR]

Our planet Earth comprises distinguished topologies based on temperature, location, latitude, longitude, and altitude, which can be captured using Remote Sensing Satellites. In this paper, the classification of satellite images is performed based on their topologies and geographical features. Researchers have worked on several machine learning and deep learning methods like support vector machine, k-nearest neighbor, maximum likelihood, deep belief network, etc. that can be used to solve satellite image classification tasks. All strategies give promising results. Recent trends show that a Convolutional Neural Network (CNN) is an excellent deep learning model for classification purposes, which is used in this paper. The open-source EuroSAT dataset is used for classifying the remote images which contain 27,000 images distributed among ten classes. The 3 baseline CNN models are pre-trained, namely- ResNet50, ResNet101, and GoogleNet models. They have other sequence layers added to them with respect to CNN, and data is pre-processed using LAB channel operations. The highest accuracy of 99.68%, precision of 99.42%, recall of 99.51%, and F- Score of 99.45% are achieved using GoogleNet over the pre-processed dataset. The proposed work is compared with the state-of-art methods and it is observed that more layers in CNN do not necessarily provide a better outcome for a medium-sized dataset. The GoogleNet, a 22-layer CNN, performs faster and better than the 50 layers CNN- ResNet50, and 101 layers CNN- ResNet101. [ABSTRACT FROM AUTHOR]

This paper proposes EMA-Net, a fully convolutional neural network, to improve the effectiveness of sea-land segmentation on Gaofen-2 images. The aim is to address the issue of low segmentation accuracy in sea-land boundary regions when using remote sensing images for sea-land segmentation. The MA-Net network structure is enhanced by splitting the EfficientNet-B0 benchmark network into five convolutional blocks. The five downsampled convolutional blocks in MA-Net are then sequentially replaced. Furthermore, an extra loss term for the sea-land boundary region is incorporated through the introduction of a boundary region enhancement loss function. This approach encourages the network to focus on learning the boundary region between the sea and land. This improves the accuracy of its prediction. The study presents the results of segmentation experiments conducted on a constructed Gaofen-2 image dataset. The improved EMA-Net model, utilizing the boundary region enhancement loss, achieves better performance than other methods for both the overall region and the sea-land boundary region. The LR (Land Recall), LP (Land Precision), SR (Sea Recall), SP (Sea Precision), F1 Score (F1-Score), mIoU (Mean Intersection over Union), and EA (Edge Accuracy) are averaged over multiple experiments to reach 97.78%, 96.63%, 97.65%, 98.48%, 97.62%, 95.37%, and 87.08% respectively. Additional experiments on the IKONOS images also confirmed the adaptability of the proposed method to high-resolution imagery. [ABSTRACT FROM AUTHOR]

Seafloor unconstrained environment video is shot in uncontrolled open sea area. There are multiple backgrounds, complex illumination and weather changes, and rapid growth of algae and attached lenses, which affect the stability of video shooting quality, resulting in difficulty in image recognition. At present, there is still no algorithm that is superior to other methods generally, and it is necessary to build a model combined with specific scenes and applications. In this paper, a fish identification method in tropical waters based on transfer learning under unconstrained environment is proposed. Firstly, the image is pre-processed by affine transformation to realize data enhancement. Furthermore, RestNet50 deep convolutional neural network is constructed based on transfer learning to compare the effect of fish recognition before and after transfer learning. The results show that, the accuracy and loss indicators are better than those of non-transfer learning when the trained model of imagenet is introduced as the initial weight of the network. When the model is trained to 150 epochs, the indicators begin to converge, which can better complete the fish identification task in tropical waters under unconstrained environment. [ABSTRACT FROM AUTHOR]

The rapid and accurate positioning of the landslides through remote sensing data plays an important role in post-disaster emergency rescue. This paper was proposed a new algorithm for landslide detection in the plateau environment. The YOLOv4 was used as the basic framework, and the MobileNetv3 model was utilized as the feature extraction network to replace the backbone neural network CSPdarknet53 which was to improve the efficiency of landslide detection. By applying depth separable convolution, the parameters of the model are decreasing significantly. To further improve the accuracy of landslide detection, the coordinate attention mechanism was introduced in the bottleneck. 3070 landslide images in the Linzhi area from 2010 to 2019 were obtained through Google Earth to train and test the model. On this basis, we compared the detection speed and accuracy of other single-stage and two-stage target detection algorithms in landslide detection. Moreover, the performances of the model were analyzed under the different attention mechanisms. The results show that our model can reduce the number of parameters by 83.59% compared with the YOLOv4 model. The accuracy of landslide detection by the model is improved to 91.2%, and the detection rate reaches 35f/s. It means that the model proposed in this study would provide useful information and rapid detection for hazard assessment and emergency rescue. [ABSTRACT FROM AUTHOR]

Curiosity of the human mind and the possibility of settlement in other planets to decrease the likelihood of human extinction have acted as a catalyst in the colonization mission of the planet Mars. Exploration, colonization and human missions to the planet are being supported by many public space agencies. Although there are several factors like toxic soil, low gravity, radiation exposures etc. that rule out the possibility of colonization, the presence of polar ice caps gives abundant hope to scientists towards making Mars habitable. Colonizing the planet also considers factors like atmosphere, soil, water content etc., and there seems to be an ongoing debate on how to make the planet habitable for mankind. In order to strengthen or weaken the claim there is a necessity to explore many other factors that may contribute to Mars' colonization in the future. Weather is one such factor worth exploring. In this paper we present some artificial intelligence techniques for analyzing Martian weather data. We rely on machine learning models like Convolution Neural Networks (CNN), Gated Recurrent Units (GRU), Long Short Term Memory (LSTM), stacked LSTM, and CNN-LSTM models to analyze the red planet's weather data. The models have been validated using statistical parameters such as MAE, MSE, RMSE and R-squared coefficient. Our analysis reports that the LSTM model outperforms all the baseline models with the R-squared value as 0.8640, and the MAE value as 0.1257. [ABSTRACT FROM AUTHOR]

The term super-resolution (SR) refers to the applications of enhancing the resolution of an image from low-resolution (LR) to high-resolution (HR). However, for compression purposes, we can intentionally down-sample the image during encoding and then up-sample it with a super-resolution technique at the decoder. Integrating this processing in a compression scheme will allow us to perform improvements through: (1) reducing the loss of information by producing a controlled down-sampling, which preserves the relevant information; (2) increasing the information compensation via a super-resolution technique that enhances the high-frequency components of the reconstructed image. In this paper, we propose a lossy hyperspectral image compression scheme. It combines a 3D wavelet transform with a wavelet learned-based super-resolution technique. The 3D wavelet transform introduces a down-sampling on the hyperspectral image to generate low-resolution (LR) images. This LR sequence is then lossy compressed by the 3D SPIHT encoder at the selected bitrate and produces the bitstream to save or transmit. At the decoder, we do the reconstruction by inverting the 3D SPIHT, which provides us with the approximate coefficients (AC) and corresponding detail coefficients (DC) of the LR image across all sub-bands. Then, we develop a wavelet learning-based SR technique that learns the mapping between the wavelet coefficients using a convolutional neural network (CNN). We use this mapping to invert the down-sampling process and predict the missing detail coefficients. Finally, The LR image and the predicted details coefficients (DC) are used to generate the high-resolution image by applying the inverse 3D wavelet transform. Therefore, we have used the down-sampling and the super-resolution technique as a mechanism to control the compression ratio (CR) and optimize the overall quality of the reconstructed image. The performance of the proposed compression scheme has been evaluated on AVIRIS hyperspectral images and compared with the main existing algorithms. Experimental results show that the proposed algorithm provides a promising performance and can generate high-quality images. [ABSTRACT FROM AUTHOR]