Search

Aiming to address the limitations of traditional channel estimation methods in time- varying channel environments, as well as the low estimation accuracy or high complexity of deep learning-based channel estimation methods, a channel estimation network based on long short-term memory structure is proposed, which consists of a bidirectional long short-term memory (BILSTM) network and a multilayer perceptron (MLP) network, namely BiLSTM-MLP. First, the BiLSTM network is used to learn the time-varying characteristics of the channel. Then, a MLP network is used to denoise and reconstruct the channel estimation. Simulation results show that the proposed channel estimation method has better performance than traditional methods, and has lower complexity and better performance compared with the same type of deep learning-based estimation methods. Furthermore, the proposed method is also robust to different pilot densities. [ABSTRACT FROM AUTHOR]

Objectives: Landslide disaster has become the first geological disaster in China. In the process of landslide movement instability, its surface will show certain deformation information, and this deformation information is an important external manifestation of landslide internal change instability, which can be used as an important reference information for landslide prediction and early warning. Therefore, the high-precision prediction of landslide displacement has important reference value for landslide disaster prediction and early warning. Methods: Considering the advantages of the intelligent optimization and machine learning algorithms, a landslide displacement combined prediction model is established. First, the multilayer perceptron (MLP) method is used to preliminary prediction of landslide displacement. Then, the artificial fish swarm algorithm (AFSA) improved based on differential evolution (DE) algorithm is further constructed, and combined with the support vector regression (SVR) to established an optimal combination SVR (OPS⁃ VR) to modify the prediction results calculated by MLP. Results: The two typical landslides monitored by BeiDou show that: (1) For HF08, the single SVR prediction model had a root mean square error (RMSE) of 23.8 mm between the prediction and the true values on the test set, and RMSE for MLP, genetic algorithm-SVR (GA-SVR), particle swarm optimization-SVR (PSO-SVR) and OPSVR prediction models are 19.9 mm, 10.9 mm, 8.3 mm and 8.3 mm, respectively. Compared with the SVR, MLP, GA-SVR, PSO-SVR and OPSVR prediction models, the prediction accuracy of MLP-OPSVR prediction model is improved by 6.2 times, 5.0 times, 2.3 times, 1.5 times and 1.5 times, respectively. The RMSE of MLP-OPSVR prediction model is only 3.3 mm. (2) For HF05, the single SVR prediction model had a RMSE of 26.5 mm between the prediction and the true values on the test set, and RMSE for MLP, GA-SVR, PSO-SVR and OPSVR prediction models are 32.8 mm, 15.4 mm, 15.6 mm and 15.5 mm, respectively. However, the RMSE of MLP-OPSVR prediction model is only 7.0 mm. Compared with the SVR, MLP, GA-SVR, PSO-SVR and OPSVR prediction models, the prediction accuracy of MLP-OPSVR prediction model is improved by 2.8 times, 3.7 times, 1.2 times, 1.2 times and 1.2 times, respectively. Conclusions: The DE algorithm can effectively overcome the outstanding problem that most artificial fish individuals are in random motion in the later stage of AFSA operation and cannot search for the global optimal solution, and improves its optimization performance. Further combined with SVR, it can more reasonably determine the super parameters of SVR and improve its prediction accuracy. Compared with a single MLP and SVR prediction model, as well as combined prediction models of conventional intelligent optimization algorithm (genetic algorithm, particle swarm optimization algorithm) and improved AFSA with SVR, the MLP-OPSVR combined prediction model has higher precision prediction results, and has a good value for popularization and application in landslide warning research. [ABSTRACT FROM AUTHOR]

To deal with the issues of low accuracy of apple leaf disease images recognition under weak supervision with only image category labeling, an improved CBAM-ResNet-based algorithm was proposed in this research. Using ResNet18 as the base model, the multilayer perceptron (MLP) in the lightweight convolutional block attention module (CBAM) attention mechanism channel was improved by up-dimensioning to amplify the details of apple leaf disease features. The improved CBAM attention module was incorporated into the residual module to enhance the key details of AlphaDropout with SeLU (Scaled Exponential Linearunits) to prevent overfitting of its network and accelerate the convergence effect of the model. Finally, the learning rate was adjusted using a single-cycle cosine annealing algorithm to obtain the disease recognition model. The training test was performed under weak supervision with only image-level annotation of all sample images, which greatly reduced the annotation cost. Through ablation experiments, the best dimensional improvement of MLP in CBAM was explored as 2. Compared with the original CBAM, the accuracy rate was increased by 0.32%, and the training time of each round was reduced by 8 s when the number of parameters increased by 17.59%. Tests were conducted on a dataset of 6185 images containing five diseases, including apple spotted leaf drop, brown spot, mosaic, gray spot, and rust, and the results showed that the model achieved an average recognition accuracy of 98.44% for the five apple diseases under weakly supervised learning. The improved CBAM-ResNet18 had increased by 1.47% compared with the pre-improved ResNet18, and was higher than VGG16, DesNet121, ResNet50, ResNeXt50, EfficientNet-B0 and Xception control model. In terms of learning efficiency, the improved CBAM-ResNet18 compared to ResNet18 reduced the training time of each round by 6 s under the condition that the number of parameters increased by 24.9%, and completed model training at the fastest speed of 137 s per round in VGG16, DesNet121, ResNet50, ResNeXt50, Efficient Net-B0 and Xception control models. Through the results of the confusion matrix, the average precision, average recall rate, and average F1 score of the model were calculated to reach 98.43%, 98.46%, and 0.9845, respectively. The results showed that the proposed improved CBAM-ResNet18 model could perform apple leaf disease identification and had good identification results, and could provide technical support for intelligent apple leaf disease identification providing.

To deal with the issues of low accuracy of apple leaf disease images recognition under weak supervision with only image category labeling, an improved CBAM-ResNet-based algorithm was proposed in this research. Using ResNet18 as the base model, the multilayer perceptron (MLP) in the lightweight convolutional block attention module (CBAM) attention mechanism channel was improved by up-dimensioning to amplify the details of apple leaf disease features. The improved CBAM attention module was incorporated into the residual module to enhance the key details of AlphaDropout with SeLU (Scaled Exponential Linearunits) to prevent overfitting of its network and accelerate the convergence effect of the model. Finally, the learning rate was adjusted using a single-cycle cosine annealing algorithm to obtain the disease recognition model. The training test was performed under weak supervision with only image-level annotation of all sample images, which greatly reduced the annotation cost. Through ablation experiments, the best dimensional improvement of MLP in CBAM was explored as 2. Compared with the original CBAM, the accuracy rate was increased by 0.32%, and the training time of each round was reduced by 8 s when the number of parameters increased by 17.59%. Tests were conducted on a dataset of 6185 images containing five diseases, including apple spotted leaf drop, brown spot, mosaic, gray spot, and rust, and the results showed that the model achieved an average recognition accuracy of 98.44% for the five apple diseases under weakly supervised learning. The improved CBAM-ResNet18 had increased by 1.47% compared with the pre-improved ResNet18, and was higher than VGG16, DesNet121, ResNet50, ResNeXt50, EfficientNet-B0 and Xception control model. In terms of learning efficiency, the improved CBAM-ResNet18 compared to ResNet18 reduced the training time of each round by 6 s under the condition that the number of parameters increased by 24.9%, and completed model training at the fastest speed of 137 s per round in VGG16, DesNet121, ResNet50, ResNeXt50, Efficient Net-B0 and Xception control models. Through the results of the confusion matrix, the average precision, average recall rate, and average F1 score of the model were calculated to reach 98.43%, 98.46%, and 0.9845, respectively. The results showed that the proposed improved CBAM-ResNet18 model could perform apple leaf disease identification and had good identification results, and could provide technical support for intelligent apple leaf disease identification providing. [ABSTRACT FROM AUTHOR]