Search

Digital pathology poses unique computational challenges, as a standard gigapixel slide may comprise tens of thousands of image tiles 1-3 . Prior models have often resorted to subsampling a small portion of tiles for each slide, thus missing the important slide-level context 4 . Here we present Prov-GigaPath, a whole-slide pathology foundation model pretrained on 1.3 billion 256 × 256 pathology image tiles in 171,189 whole slides from Providence, a large US health network comprising 28 cancer centres. The slides originated from more than 30,000 patients covering 31 major tissue types. To pretrain Prov-GigaPath, we propose GigaPath, a novel vision transformer architecture for pretraining gigapixel pathology slides. To scale GigaPath for slide-level learning with tens of thousands of image tiles, GigaPath adapts the newly developed LongNet 5 method to digital pathology. To evaluate Prov-GigaPath, we construct a digital pathology benchmark comprising 9 cancer subtyping tasks and 17 pathomics tasks, using both Providence and TCGA data 6 . With large-scale pretraining and ultra-large-context modelling, Prov-GigaPath attains state-of-the-art performance on 25 out of 26 tasks, with significant improvement over the second-best method on 18 tasks. We further demonstrate the potential of Prov-GigaPath on vision-language pretraining for pathology 7,8 by incorporating the pathology reports. In sum, Prov-GigaPath is an open-weight foundation model that achieves state-of-the-art performance on various digital pathology tasks, demonstrating the importance of real-world data and whole-slide modelling., (© 2024. The Author(s).)

Neuromorphic data, recording frameless spike events, have attracted considerable attention for the spatiotemporal information components and the event-driven processing fashion. Spiking neural networks (SNNs) represent a family of event-driven models with spatiotemporal dynamics for neuromorphic computing, which are widely benchmarked on neuromorphic data. Interestingly, researchers in the machine learning community can argue that recurrent (artificial) neural networks (RNNs) also have the capability to extract spatiotemporal features although they are not event-driven. Thus, the question of "what will happen if we benchmark these two kinds of models together on neuromorphic data" comes out but remains unclear. In this work, we make a systematic study to compare SNNs and RNNs on neuromorphic data, taking the vision datasets as a case study. First, we identify the similarities and differences between SNNs and RNNs (including the vanilla RNNs and LSTM) from the modeling and learning perspectives. To improve comparability and fairness, we unify the supervised learning algorithm based on backpropagation through time (BPTT), the loss function exploiting the outputs at all timesteps, the network structure with stacked fully-connected or convolutional layers, and the hyper-parameters during training. Especially, given the mainstream loss function used in RNNs, we modify it inspired by the rate coding scheme to approach that of SNNs. Furthermore, we tune the temporal resolution of datasets to test model robustness and generalization. At last, a series of contrast experiments are conducted on two types of neuromorphic datasets: DVS-converted (N-MNIST) and DVS-captured (DVS Gesture). Extensive insights regarding recognition accuracy, feature extraction, temporal resolution and contrast, learning generalization, computational complexity and parameter volume are provided, which are beneficial for the model selection on different workloads and even for the invention of novel neural models in the future., Competing Interests: Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2020 Elsevier Ltd. All rights reserved.)

Next‐generation light source facilities offer extreme spatial and temporal resolving power, enabling multiscale, ultra‐fast and dynamic characterizations. However, a trade‐off between acquisition efficiency and data quality needs to be made to fully unleash the resolving potential, for which purpose powerful denoising algorithms to improve the signal‐to‐noise ratio of the acquired X‐ray images are desirable. Yet, existing models based on machine learning mostly require massive and diverse labeled training data. Here we introduce a self‐supervised pre‐training algorithm with blind denoising capability by exploring the intrinsic physical symmetry of X‐ray patterns without requiring high signal‐to‐noise ratio reference data. The algorithm is more efficient and effective than algorithms without symmetry involved, including an supervised algorithm. It allows us to recover physical information from spatially and temporally resolved data acquired in X‐ray diffraction/scattering and pair distribution function experiments, where pattern symmetry is often well preserved. This study facilitates photon‐hungry experiments as well as in situ experiments with dynamic loading. [ABSTRACT FROM AUTHOR]

Leaf nitrogen concentration (LNC) is a primary indicator of crop nitrogen status, closely related to the growth and development dynamics of crops. Accurate and efficient monitoring of LNC is significant for precision field crop management and enhancing crop productivity. However, the biochemical properties and canopy structure of wheat change across different growth stages, leading to variations in spectral responses that significantly impact the estimation of wheat LNC. This study aims to investigate the construction of feature combination indices (FCIs) sensitive to LNC across multiple wheat growth stages, using remote sensing data to develop an LNC estimation model that is suitable for multiple growth stages. The research employs UAV multispectral remote sensing technology to acquire canopy imagery of wheat during the early (Jointing stage and Booting stage) and late (Early filling and Late filling stages) in 2021 and 2022, extracting spectral band reflectance and texture metrics. Initially, twelve sensitive spectral feature combination indices (SFCIs) were constructed using spectral band information. Subsequently, sensitive texture feature combination indices (TFCIs) were created using texture metrics as an alternative to spectral bands. Machine learning algorithms, including partial least squares regression (PLSR), random forest regression (RFR), support vector regression (SVR), and Gaussian process regression (GPR), were used to integrate spectral and texture information, enhancing the estimation performance of wheat LNC across growth stages. Results show that the combination of Red, Red edge, and Near-infrared bands, along with texture metrics such as Mean, Correlation, Contrast, and Dissimilarity, has significant potential for LNC estimation. The constructed SFCIs and TFCIs both enhanced the responsiveness to LNC across multiple growth stages. Additionally, a sensitive index, the Modified Vegetation Index (MVI), demonstrated significant improvement over NDVI, correcting the over-saturation concerns of NDVI in time-series analysis and displaying outstanding potential for LNC estimation. Spectral information outperforms texture information in estimation capability, and their integration, particularly with SVR, achieves the highest precision (coefficient of determination (R2) = 0.786, root mean square error (RMSE) = 0.589%, and relative prediction deviation (RPD) = 2.162). In conclusion, the sensitive FCIs developed in this study improve LNC estimation performance across multiple growth stages, enabling precise monitoring of wheat LNC. This research provides insights and technical support for the construction of sensitive indices and the precise management of nitrogen nutrition status in field crops. [ABSTRACT FROM AUTHOR]

Unmanned aerial vehicle (UAV) offers an unprecedented observing potential with ultrahigh spatial and temporal resolutions and high flexibility. However, it remains difficult to solve the radiometric inconsistency of multiflight UAV imagery. This study proposed a relative radiometric correction method for multiflight UAV images based on concurrent satellite imagery (MACA) consisting of two steps, i.e., cross-sensor spectral fitting (CSF) and fine-resolution spectral calibration (FSC). In CSF, relationships of multiflight UAV reflectance and the concurrent satellite reflectance were established for generating the fine-resolution reference imagery. Subsequently, a relative radiometric correction model was constructed in the FSC step to correct multiflight UAV imagery. The performance of MACA was evaluated using multiflight UAV datasets acquired on six cropland sites and a concurrent Sentinel-2 image. Compared with four typical or state-of-the-art relative correction methods, the correction using MACA yielded better consistency between UAV and Sentinel-2 data, regardless of individual spectral bands ($\text{R}^{2} =0.79$ –0.86, root mean square error (RMSE) $=0.004$ –0.019) or vegetation indices (VIs) ($\text{R}^{2} =0.80$ –0.86, RMSE $=0.024$ –0.054). Moreover, the prediction of plant nitrogen accumulation (PNA) based on the MACA-corrected UAV data had the highest accuracy and showed the spatial variation most significantly within and between fields for all sites. The results demonstrated that MACA was more robust in reducing spectral mismatch across sensors and eliminating the subjective error of pseudo-invariant features (PIFs) selection. MACA has the potential to be used to cross-calibrate multisensor data into a consistent standard, which will benefit multisensor synergies. [ABSTRACT FROM AUTHOR]

In the future, big data will become an efficient and useful means for improving urban planning, and machine learning can take city as a simplified and efficient system. We take full advantage of the benefits of new technology, but also clarify that city is not a machine, also cannot fully mechanically control the urban development. This study presents a methodology for identifying low-carbon travel block, which can be used to identify the built environment conducive to residents' low-carbon travel. We chose the four elements of traffic survey—travel mode, travel time, travel purpose and travel frequency—as the framework to evaluate travel carbon emissions. Using the index data collected from "WeChat," a popular social-media platform in China and questionnaire surveys, we conducted hotspot analysis of the spatial distribution of travel carbon emissions in GIS. We obtained a comprehensive carbon emissions and its spatial distribution through the superposition of hotspot density surface of different indexes. The results show that E block within the research area has the lowest travel carbon emissions. These results suggest some planning implications from three aspects—land use mode, road network and public service facilities: In the old urban district of Pucheng, the ratio of residential building area and other types' building area should be "4:1–3:1"; and we should develop the travel model of bicycle, and the interval of bicycle lanes should be 350–450 m; The ratio of walking road to total road area should be 15–20%, and the width of road should be restricted. Coverage of transit site buffered for the radius of 150 m is 40–50%, coverage of shopping services buffered for the radius of 50 m is 45–60%, and coverage of recreational facilities buffered for the radius of 100 m is 50–70%. The results confirm that "functional mixing" and "dense road network" are beneficial to residents' low-carbon travel proposed by the predecessors. At the same time, we found that not the higher volume rate is, the more favorable for low-carbon travel. Small cities have limited number of population and scattered distribution of professional posts, which are not suitable for the traditional mode of improving the volume ratio and the bus system. It is not that the higher the bus station coverage is, the better for residents to travel as low-carbon, and the high popularity of public transportation in small cities will increase the carbon emission of residents. The study provides a new way to evaluate the carbon emission assessment of blocks and provides a basis for block planning with low-carbon concept. [ABSTRACT FROM AUTHOR]

• A decomposition-integrated model was proposed to predict influent NH 3 -N of WWTP. • Rolling decomposition method ensures an information leakage-free modeling process. • The proposed integrated model shows better performance compared with single model. • The model outperforms the integrated models with information leakage in practice. Timely and accurate assessment of key sewage quality indicators based on deep learning models has attracted much attention for intelligent wastewater treatment. Decomposition algorithms are widely adopted to further enhance the prediction performance of deep learning models, but traditional one-time decomposition method suffers from information leakage. Herein, an information leakage-free integrated model enabled by rolling decomposition method was built to predict influent ammonia nitrogen (NH 3 -N), a key indicator for wastewater treatment. Firstly, the original NH 3 -N sequence is decomposed into different subsequences using variational mode decomposition (VMD) under rolling method, which adds data successively for decomposition and excludes the future data, thus avoiding information leakage. Secondly, model and predict the subsequences by gated recurrent unit (GRU). Finally, summing the prediction results of the subsequences to obtain the prediction result of NH 3 -N. The results show that the proposed model outperforms single GRU, with 16.69% lower RMSE, 13.02% lower MAE, and 11.90% lower MAPE. Moreover, the model also has advantages over the integrated model that trained under information leakage, with 42.34% lower RMSE, 41.06% lower MAE, and 39.34% lower MAPE. This work improves the applicability of integrated models and assists in the intelligent wastewater treatment and the construction of sustainable cities. [ABSTRACT FROM AUTHOR]

Human activities and climate change are constantly threatening water environment systems. To alleviate pressure on the water environment and meet the requirements for sustainable societal development, it has become important to study the water environment carrying capacity (WECC). As science and technology have advanced, the use of artificial intelligence has penetrated various disciplines, and the application of machine learning has achieved many good results in the environmental sciences; however, little attention has been given to the WECC. To this end, a WECC assessment model was developed based on a variety of machine learning algorithms, to provide a novel approach and better understanding of the WECC. In this study, 13 natural and anthropogenic factors related to the water environment system in Liaoning Province (China) were selected, and the study period ranged from 2015 to 2019. Four basic machine learning algorithms, namely, random forest (RF), support vector machine (SVM), multilayer perceptron (MLP) and k-nearest neighbour (KNN), were used for batch modelling, and stacking was carried out based on the output results. A comparison and screening of the models revealed that RF and the stacking model had the best model prediction performances. Because RF was relatively simple and convenient, a comprehensive assessment of the regional WECC based on this algorithm was carried out. The results showed that most of the control units with an overloaded WECC are located in the central and northern regions of Liaoning Province, and the control units of overloading can be well identified and the trend of regional WECC status can be observed. To improve model transparency, we also performed an explanatory analysis based on the shapley additive explanations (SHAP). The results showed that precipitation, soil texture, sewage treatment plant, wind speed, and temperature were the most important factors influencing the water environment system. Precipitation is the most influential factor, and with increasing rainfall, the WECC first improves and then gradually deteriorates. For the anthropogenic factors, the WECC generally improves first and then worsens as these factors increase. These research results can be used to provide a powerful and practical new methodology for the comprehensive assessment of the WECC and provide a basis for the development of water environment policies and regulatory implementation in the region. [Display omitted] • A regional WECC assessment method based on machine learning was developed. • Batch modelling and model stacking were performed via machine learning. • Model explanation was based on the SHAP analysis method. • Results of the assessment are critical to the management of water environmental systems. • Machine learning is a promising method for assessing the WECC. [ABSTRACT FROM AUTHOR]

[Display omitted] • Propose deep learning-based detection method of time-delayed water quality indicator. • Contribution of each input indicator is quantified based on SHAP value. • NH 3 -H, DO, turbidity contribute significantly to detecting time-delayed WQIs. • Sensitivity analysis provides insight for higher accuracy and cost performance. Rapid and accurate detection of time-delayed water quality indicators (WQIs) is the key to achieving fast feedback regulation of wastewater treatment plants (WWTPs) that enables its energy-efficient operation and high tolerance towards shock sewage loads. However, advanced oxidation methods are costly, and data-driven modeling methods based on traditional machine learning algorithms for detecting time-delayed WQIs have limited detection accuracy. This work develops deep learning models based on long short-term memory (LSTM) neural networks to detect time-delayed WQIs in WWTPs intake accurately. The lack of interpretability of the deep learning models hampers the optimization of the developed LSTM models in applications. Therefore, a global sensitivity analysis (GSA) based on Shapley additive explanations (SHAP) is performed to quantify the contribution of the input indicators to detection results of the developed LSTM models. The direct contributions provide the basis for optimizing the input indicators to achieve more cost-effective modeling detection. In the case study, the developed LSTM models achieved good accuracy (R2 of 0.9141, 0.9239, and 0.9040, respectively) in detecting chemical oxygen demand, total nitrogen, and total phosphorus in the influent of a WWTP, outperforming the four types of baseline models. According to the SHAP values, the contributions of dissolved oxygen, turbidity, and ammonia nitrogen to the above detection targets are always in the top third of all input indicators, which are more outstanding than meteorological indicators. Removing the indicator with the smallest SHAP value reduces the build and run costs of the models with minimal loss of detection accuracy. Combining deep learning and GSA to detect WWTPs influent is a novel and effective attempt. This attempt provides a more sustainable solution for rapid and accurate detection of time-delayed WQIs, which drives WWTPs' operation in an intelligent, clean, and safe direction. [ABSTRACT FROM AUTHOR]

• A model combining two-stage decomposition and LSTM for TP detection in sewer networks. • Two-stage decomposition reduces the instability and complexity of TP time series. • Mode decomposition procedure setting significant impact modeling efficiency, accuracy. • Proposed hybrid model achieves excellent accuracy of average R2 = 0.946. • Proposed hybrid model has the advantage of being energy-efficient and cost-effective. Timely and accurately grasping total phosphorus (TP) concentration in sewer networks is crucial for urban phosphorus flow management and shock load early warning of sewage treatment facilities. Modeling-based methods are cleaner and more energy-saving than traditional methods requiring digestion procedures. However, the TP time series' strong nonlinearity and complexity result in unsatisfactory accuracy in these methods. This work proposes a hybrid model named CEEMDAN-SE-VMD-LSTM that combines complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), sample entropy (SE), variational mode decomposition (VMD), and long short-term memory (LSTM) neural network to improve the accuracies of modeling-based methods. In proposed hybrid model, the two-stage decomposition procedure can decompose the TP time series into several lower-complexity modes, and the powerful nonlinear mapping ability of the LSTM neural network enables accurate prediction of these modes. In case study, the proposed hybrid model achieves excellent detection accuracy with an average R2 of 0.9460 ± 0.0243. Compared with the hybrid models formed by combining other decomposition procedures (i.e., CEEMDAN, VMD, singular spectrum analysis (SSA), CEEMDAN-SE-SSA) and LSTM neural network, the proposed hybrid model has the highest detection accuracy (1.36–3.94 % higher Nash-Sutcliffe efficiency, 1.28–4.58 % higher R2, 12.14–24.86 % lower RMSE). The strategy of setting the mode decomposition procedure based on a comprehensive analysis of the decomposition algorithms and the obtained modes ensures high detection accuracy of the proposed hybrid model while avoiding costly computational burdens. This work is enlightening for improving the accuracy and modeling efficiency of soft detection methods, which are expected to reduce energy consumption and pollution caused by water quality detection. [ABSTRACT FROM AUTHOR]

• Data-driven method based on deep learning can predict FOG content in sewers. • The FOG prediction model based on GRU has good generalization ability. • GRU network structure adjustment achieves the optimal model performance. • The key indicators are identified based on GSA for rapid FOG content monitoring. The content of fat, oil and grease (FOG) in the sewer network sediments is the key indicator for diagnosing sewer blockage and overflow. However, the traditional FOG detection is time-consuming and costly, and the establishment of mathematical models based on statistical methods to predict the content of FOG fail to provide satisfactory accuracy. Herein, a deep learning algorithm used a data-driven FOG content prediction model is proposed to achieve a more accurate prediction of FOG content. Meanwhile, global sensitivity analysis (GSA) is exploited to evaluate the contribution of input indicators to the output indicator (FOG) in the model, so that some input indicators that have less impact on the prediction performance can be screened out, the best combination of input indicators can be determined, and the operation cost of the model can be reduced. To evaluate the effectiveness of the proposed model, a case study was conducted in a city in southern China. The experimental results indicate that the prediction model obtains good FOG estimations and performs well from a single site to multiple sites with a mean R2 of 0.922, showing a good generalization performance. Through GSA, the key input indicators in the model were identified as pH, water temperature (T), relative humidity (RH), sewage flow (Flow), drinking water supply (DWS), velocity (V) and conductivity (σ), and the input indicators such as air pressure (AP), population (Pop.), and liquid level (LV) can be reduced without affecting the prediction accuracy of the model. [Display omitted] [ABSTRACT FROM AUTHOR]

Point source pollution in urban drainage networks, which is difficult to monitor and control, has been regarded as an intractable problem. To solve the problem, key water quality indicators must be tracked in the evaluation and prediction of sewer water quality. However, some of these important chemical indicators (e.g. biological oxygen demand (BOD 5), chemical oxygen demand (COD), ammonia nitrogen (NH 4 +-N), total nitrogen (TN), and total phosphorus (TP)) require a great deal of time and effort to measure, which will adversely affect the prediction in a sewage network. Existing statistical methods and machine learning algorithms cannot effectively solve the detection time problem or provide limited accuracy. Moreover, the lack of various factors taken into account in these methods results in unsatisfactory predictive performance. Few studies consider the impact of urban multi-source data on water quality prediction of sewer networks while developing statistical methods or machine learning algorithms. To address this problem, we propose a deep learning approach based on multi-source data fusion. This approach takes into account the following indicators to comprehensively analyze and predict drainage water quality: environmental indicators (such as area and diameter); social indicators (such as population); water quantity indicators (such as drinking water supply, sewage flow, water velocity, and liquid level); and easily monitored water quality criteria indicators (such as pH, temperature, and conductivity). To test the effectiveness of this method, we conducted a case study in a city in southern China. By comparing this new method with the linear method (multiple linear regression, MLR) and traditional learning algorithm (multilayer perception, MLP), it is found that the deep learning algorithm—which includes recurrent neural network (RNN), long-short term memory (LSTM), and gated recurrent unit (GRU)—has good predictive performance, in which GRU shows superior ability in predicting the chemical index of water quality and the learning curve is faster. The results showed that the GRU achieved 0.82%–5.07% higher R2 than RNN and LSTM, 9.13%–15.03% higher R2 than traditional machine learning algorithms, and 37.26%–43.38% higher R2 than linear methods. [Display omitted] • Deep learning model based on multi-source data can predict sewer water quality. • RNN-based models have better predictive performance than MLR and MLP models. • In deep learning models, the GRU has fast learning curve and simple architecture. • GRU model prediction performance outperformed RNN and LSTM model. [ABSTRACT FROM AUTHOR]