A Spatial Feature-Enhanced Attention Neural Network with High-Order Pooling Representation for Application in Pest and Disease Recognition

Agriculture, Journal Year: 2022, Volume and Issue: 12(4), P. 500 - 500

Published: March 31, 2022

With the development of advanced information and intelligence technologies, precision agriculture has become an effective solution to monitor prevent crop pests diseases. However, pest disease recognition in applications is essentially fine-grained image classification task, which aims learn discriminative features that can identify subtle differences among similar visual samples. It still challenging solve for existing standard models troubled by oversized parameters low accuracy performance. Therefore, this paper, we propose a feature-enhanced attention neural network (Fe-Net) handle diseases innovative agronomy practices. This model established based on improved CSP-stage backbone network, offers massive channel-shuffled various dimensions sizes. Then, spatial module added exploit interrelationship between different semantic regions. Finally, proposed Fe-Net employs higher-order pooling mine more highly representative computing square root covariance matrix elements. The whole architecture efficiently trained end-to-end way without additional manipulation. comparative experiments CropDP-181 Dataset, achieves Top-1 Accuracy up 85.29% with average time only 71 ms, outperforming other methods. More experimental evidence demonstrates our approach obtains balance model’s performance parameters, suitable its practical deployment art applications.

Language: Английский

---

A Reversible Automatic Selection Normalization (RASN) Deep Network for Predicting in the Smart Agriculture System

Agronomy, Journal Year: 2022, Volume and Issue: 12(3), P. 591 - 591

Published: Feb. 27, 2022

Due to the nonlinear modeling capabilities, deep learning prediction networks have become widely used for smart agriculture. Because sensing data has noise and complex nonlinearity, it is still an open topic improve its performance. This paper proposes a Reversible Automatic Selection Normalization (RASN) network, integrating normalization renormalization layer evaluate select module of model. The accuracy been improved effectively by scaling translating input with learnable parameters. application results show that model good ability adaptability greenhouse in Smart Agriculture System.

Language: Английский

---

A Variational Bayesian Deep Network with Data Self-Screening Layer for Massive Time-Series Data Forecasting

Entropy, Journal Year: 2022, Volume and Issue: 24(3), P. 335 - 335

Published: Feb. 25, 2022

Compared with mechanism-based modeling methods, data-driven based on big data has become a popular research field in recent years because of its applicability. However, it is not always better to have more when building forecasting model practical areas. Due the noise and conflict, redundancy, inconsistency time-series data, accuracy may reduce contrary. This paper proposes deep network by selecting understanding improve performance. Firstly, self-screening layer (DSSL) maximal information distance coefficient (MIDC) designed filter input high correlation low redundancy; then, variational Bayesian gated recurrent unit (VBGRU) used anti-noise ability robustness model. Beijing's air quality meteorological are conducted verification experiment 24 h PM2.5 concentration forecasting, proving that proposed superior other models accuracy.

Language: Английский

---

Ada-STGMAT: An adaptive spatio-temporal graph multi-attention network for intelligent time series forecasting in smart cities

Expert Systems with Applications, Journal Year: 2025, Volume and Issue: 269, P. 126428 - 126428

Published: Jan. 10, 2025

Language: Английский

---

Overall recursive least squares and overall stochastic gradient algorithms and their convergence for feedback nonlinear controlled autoregressive systems

International Journal of Robust and Nonlinear Control, Journal Year: 2022, Volume and Issue: 32(9), P. 5534 - 5554

Published: April 4, 2022

Abstract This article deals with the problems of parameter estimation for feedback nonlinear controlled autoregressive systems (i.e., equation‐error systems). The bilinear‐in‐parameter identification model is formulated to describe system. An overall recursive least squares algorithm developed handle difficulty bilinear‐in‐parameter. For purpose avoiding heavy computational burden, an stochastic gradient deduced and forgetting factor introduced improve convergence rate. Furthermore, analysis proposed algorithms are established by means process theory. effectiveness illustrated simulation example.

Language: Английский

---

A novel multi-innovation gradient support vector machine regression method

ISA Transactions, Journal Year: 2022, Volume and Issue: 130, P. 343 - 359

Published: March 17, 2022

Language: Английский

---

BMAE-Net: A Data-Driven Weather Prediction Network for Smart Agriculture

Agronomy, Journal Year: 2023, Volume and Issue: 13(3), P. 625 - 625

Published: Feb. 22, 2023

Weather is an essential component of natural resources that affects agricultural production and plays a decisive role in deciding the type production, planting structure, crop quality, etc. In field agriculture, medium- long-term predictions temperature humidity are vital for guiding activities improving yield quality. However, existing intelligent models still have difficulties dealing with big weather data predicting applications, such as striking balance between prediction accuracy learning efficiency. Therefore, multi-head attention encoder-decoder neural network optimized via Bayesian inference strategy (BMAE-Net) proposed herein to predict time series changes accurately. Firstly, we incorporate into gated recurrent unit construct Bayesian-gated units (Bayesian-GRU) module. Then, mechanism introduced design structure each layer, applicability time-length changes. Subsequently, framework hyperparameter optimization designed infer intrinsic relationships among time-series high accuracy. For example, R-evaluation metrics three locations 0.9, 0.804, 0.892, respectively, while RMSE reduced 2.899, 3.011, 1.476, seen Case 1 data. Extensive experiments subsequently demonstrated BMAE-Net has overperformed on location datasets, which provides effective solution applications smart agriculture system.

Language: Английский

---

Deep learning and gradient boosting for urban environmental noise monitoring in smart cities

Expert Systems with Applications, Journal Year: 2023, Volume and Issue: 218, P. 119568 - 119568

Published: Jan. 20, 2023

Language: Английский

---

An Effective Pyramid Neural Network Based on Graph-Related Attentions Structure for Fine-Grained Disease and Pest Identification in Intelligent Agriculture

Agriculture, Journal Year: 2023, Volume and Issue: 13(3), P. 567 - 567

Published: Feb. 26, 2023

In modern agriculture and environmental protection, effective identification of crop diseases pests is very important for intelligent management systems mobile computing application. However, the existing mainly relies on machine learning deep networks to carry out coarse-grained classification large-scale parameters complex structure fitting, which lacks ability in identifying fine-grained features inherent correlation mine pests. To solve problems, a pest method based graph pyramid attention, convolutional neural network (GPA-Net) proposed promote agricultural production efficiency. Firstly, CSP backbone constructed obtain rich feature maps. Then, cross-stage trilinear attention module extract abundant discrimination portions objects as much possible. Moreover, multilevel designed learn multiscale spatial graphic relations enhance recognize diseases. Finally, comparative experiments executed cassava leaf, AI Challenger, IP102 datasets demonstrates that GPA-Net achieves better performance than models, with accuracy up 99.0%, 97.0%, 56.9%, respectively, more conducive distinguish applications practical smart protection.

Language: Английский

---

A Graph-Related High-Order Neural Network Architecture via Feature Aggregation Enhancement for Identification Application of Diseases and Pests

Computational Intelligence and Neuroscience, Journal Year: 2022, Volume and Issue: 2022, P. 1 - 16

Published: May 26, 2022

Diseases and pests are essential threat factors that affect agricultural production, food security supply, ecological plant diversity. However, the accurate recognition of various diseases is still challenging for existing advanced information intelligence technologies. Disease pest typically a fine-grained visual classification problem, which easy to confuse traditional coarse-grained methods due external similarity between different categories significant differences among each subsample same category. Toward this end, paper proposes an effective graph-related high-order network with feature aggregation enhancement (GHA-Net) handle image diseases. In our approach, improved CSP-stage backbone first formed offer massive channel-shuffled features in multiple granularities. Secondly, relying on multilevel attention mechanism, module designed exploit distinguishable representing discriminating parts. Meanwhile, graphic convolution constructed analyse graph-correlated representation part-specific interrelationships by regularizing semantic into tensor space. With collaborative learning three modules, approach can grasp robust contextual details better identification. Extensive experiments several public disease datasets demonstrate proposed GHA-Net achieves performances accuracy efficiency surpassing other models more suitable identification applications complex scenes.

Language: Английский

---