The traditional Deep Learning （DL）-based multi-objective solvers have the problems of low model utilization and being easy to fall into the local optimum. Aiming at these problems， a Multi-objective Optimization model for Unmanned aerial vehicles Trajectory based on Decomposition and Trajectory search （DTMO-UT） was proposed. The proposed model consists of the encoding and decoding parts. First， a Device encoder （Dencoder） and a Weight encoder （Wencoder） were contained in the encoding part， which were used to extract the state information of the Internet of Things （IoT） devices and the features of the weight vectors. And the scalar optimization sub-problems that were decomposed from the Multi-objective Optimization Problem （MOP） were represented by the weight vectors. Hence， the MOP was able to be solved by solving all the sub-problems. The Wencoder was able to encode all sub-problems， which improved the utilization of the model. Then， the decoding part containing the Trajectory decoder （Tdecoder） was used to decode the encoding features to generate the Pareto optimal solutions. Finally， to alleviate the phenomenon of greedy strategy falling into the local optimum， the trajectory search technology was added in trajectory decoder， that was generating multiple candidate trajectories and selecting the one with the best scalar value as the Pareto optimal solution. In this way， the exploration ability of the trajectory decoder was enhanced during trajectory planning， and a better-quality Pareto set was found. The results of simulation experiments show that compared with the mainstream DL MOP solvers， under the condition of 98.93% model parameter quantities decreasing， the proposed model reduces the distribution of MOP solutions by 0.076%， improves the ductility of the solutions by 0.014% and increases the overall performance by 1.23%， showing strong ability of practical trajectory planning of DTMO-UT model.
In view of the problems of the current Convolutional Neural Network （CNN） using end layer features to recognize facial expression， such as complex model structure， too many parameters and unsatisfactory recognition， an optimization algorithm based on the combination of improved CNN and Support Vector Machine （SVM） was proposed. First， the network model was designed by the idea of continuous convolution to obtain more nonlinear activations. Then， the adaptive Global Average Pooling （GAP） layer was used to replace the fully connected layer in traditional CNN to reduce the network parameters. Finally， in order to improve generalization ability of the model， SVM classifier instead of the traditional Softmax function was used to realize expression recognition. Experimental results show that the proposed algorithm achieves 73.4% and 98.06% recognition accuracy on Fer2013 and CK+ datasets， which is 2.2 percentage points higher than the traditional LeNet-5 algorithm on Fer2013 dataset. Moreover， this network model has simple structure， less parameters and good robustness.
Aiming at the issue of ineffective fusion of multi-modal features of indoor scene semantic segmentation using RGB-D， a network named APFNet （Attention mechanism and Pyramid Fusion Network） was proposed， in which attention mechanism fusion module and pyramid fusion module were designed. To fully use the complementarity of the RGB features and the Depth features， the attention allocation weights of these two kinds of features were respectively extracted by the attention mechanism fusion module， making the network focus more on the multi-modal feature domain with more information content. Local and global information were fused by pyramid fusion module with four different scales of pyramid features， thus scene context was extracted and segmentation accuracies of object edges and small-scale objects were improved. By integrating these two fusion modules into a three-branch “encoder-decoder” network， an “end-to-end” output was realized. Comarative experiments were implemented with the state-of-the-art methods， such as multi-level RGB-D residual feature Fusion network （RDF-152）， Attention Complementary features Network （ACNet） and Spatial information Guided convolution Network （SGNet） on the SUN RGB-D and NYU Depth v2 datasets. Compared with the best-performing method RDF-152， when the layer number of the encoder network was reduced from 152 to 50， the Pixel Accuracy （PA）， Mean Pixel Accuracy （MPA）， and Mean Intersection over Union （MIoU） of APFNet were respectively increased by 0.4， 1.1 and 3.2 percentage points. The semantic segmentation accuracies for small-scale objects such as pillows and photos， and large-scale objects such as boards and ceilings were increased by 0.9 to 3.4 and 12.4 to 18 percentage points respectively. The results show that the proposed APFNet has some advantages in dealing with the semantic segmentation of indoor scenes.
To solve the problem of Fine Particulate Matter (PM2.5) concentration prediction, a PM2.5 concentration prediction model was proposed. First, through introducing the comprehensive meteorological index, the factors of wind, humidity, temperature were comprehensively considered; then the feature vector was conducted by combining the actual concentration of SO2, NO2, CO and PM10; finally the Least Squares Support Vector Machine (LS-SVM) prediction model was built based on feature vector and PM2.5 concentration data. The experimental results using the data from the city A and city B environmental monitoring centers in 2013 show that, the forecast accuracy is improved after the introduction of a comprehensive weather index, error is reduced by nearly 30%. The proposed model can more accurately predict the PM2.5 concentration and it has a high generalization ability. Furthermore, the author analyzed the relationship between PM2.5 concentration and the rate of hospitalization, hospital outpatient service amount, and found a high correlation between them.
