The traditional text summarization generation model based on sequence does not consider the context semantic information of words, which leads to the low accuracy of the generated text summary and does not conform to the human language habits. In view of this inadequacy, a structure-based attention sequence to sequence model (SBA) is proposed, which combines the sequence to sequence generation model of attention mechanism and introduces the syntactic structure information of the text, so that the context vector obtained by the attention structure contains both the semantic information and the syntactic structure information of the text, so as to obtain the generated text abstract. The experimental results based on Gigaword dataset show that the proposed method can effectively improve the accuracy and readability of the generated summary.

Automated planning is an important branch of artificial intelligence, which can be widely used in robot, industrial production, and commercial fields. A review is conducted on the temporal planning, which is a frontier and sub-area of automated planning, from three aspects: 1) temporal feature; 2) planning method; 3) applications. The conclusions are: 1) comparing with planning ability, the community has proposed sufficient temporal features; 2) for the time being, state space forward searching based on heuristics is our best choice; 3) researchers are still looking for more and better applications. This survey aims to provide basic concepts for students who are interested in temporal planning.

Based on the Guangzhou license plate auction price data set, linear regression combined with k-fold cross-validation is used to study the prediction method of a small sample data set. In order to solve the problem of increased verification errors caused by local specific data in a small sample set, a strategy to shuffle the data globally before verification is proposed. Finally, it is verified through experiments that this strategy can significantly reduce the verification error. Based on this, through multiple sets of experimental verification, the appropriate parameters are determined, and the results show that the total average correct rate of the final predicted value has reached 95%.

A novel semi-supervised two-view multi-instance clustering model is proposed, which bands text-view with image-view and solves the multi-instance image clustering problem with a small amount of label. Firstly, the proposed model embeds Concept Factorization and multi-instance kernel into a joint framework, which learns the association matrix of each view and the cluster indicator matrix shared by both views. Then, a ${l_{2, 1}}$-norm is applied to learn the optimal association matrix and cluster indicator matrix. Furthermore, to enhance the discriminability between bags, the proposed model enforces the similarity of the cluster indicators for the bag with the same label to approximate 1 and the similarity with different labels to 0. Finally, an iterative updating algorithm is derived to solve the proposed model. The experimental results show that the proposed model is superior to other multi-instance clustering models.

Intelligent agriculture has become a major trend in the development of modern agriculture. Low-altitude UAV remote sensing image analysis is the key point in precision agriculture, which guides UAV to carry out precise job by studying on hyperspectral remote sensing images taken by UAV. However, small and medium-sized farms have the disadvantage of insufficient computational resource. To deal with the problems, a classification method based on lightweight Convolutional Neural Networks (CNN) is proposed. This method aims to reduce the training cost and maintain an acceptable classification accuracy, so as to help farms avoid replacing expensive high-performance equipment. PCA is used to reduce the image spectral dimension and some data augmentation methods are used to enlarge the dataset. The CNN structure is optimized to accelerate the training process and the model used to extract Spectral-Spatial feature to improve classification accuracy. Experiments are conducted on three benchmark datasets, which shows that our lightweight CNN model can guarantee a satisfying accuracy while having a lower training cost compared with both traditional methods and deep learning method.

High-efficiency video coding (High Efficiency Video Coding, HEVC) inter-prediction part has a high computational complexity. In order to reduce the coding time complexity, a fast algorithm for inter-prediction is proposed based on depth characteristics. First, the inter-frame PU (Prediction Unit, PU) mode is classified into regular and irregular blocks; then the method is used to define the content complexity of the current CTU (Coding Tree Unit, CTU); finally, the coded depth is divided into SD and HD regions. The SD region is only divided into rules. According to the complexity of the current CTU, the HD region is used to traverse the complex blocks in an irregular manner, and the simple blocks are divided into rules to achieve a fast encoding algorithm. Experimental results show that compared with HM16.9, the proposed algorithm reduces the encoding time, respectively, by 28.6% and 23.8% under low latency and random access configurations and the encoding performance is unaffected.

As an important device for displacement, velocity and acceleration measurement, gratings are widely used in high-precision, high-speed automated CNC feedback closed-loop systems. In order to enhance the reliability of high-speed measurement, a linear encoder with I-Q balanced photoelectric coding and decoding method is proposed. The main grating and the indicating grating are placed in parallel, and the difference in grating constant forms a moiré fringe. Each photodiode of the photodetection unit is an independent detection channel, which is distributed according to the relative phase position of the moiré fringe at a fixed spatial frequency, and is amplified by an independent DC operational amplifier circuit designed according to the I-Q balanced encoding and decoding equation. I-Q balanced encoding and decoding equation conditions are respectively sent to the 2-way adder, and the output of the adder is the standard and general 2-way I-Q quadrature signal, which is in line with the standard for automation equipment. This method comprehensively enhances the automatic numerical control closed-loop system's anti-grating surface pollution, anti-photodiode distribution position error and anti-low-frequency vibration ability, and is an efficient, reliable, and practical grating technology.

The visual odometry plays a key role in the positioning and navigation of mobile robots, but the current algorithm still has room for improvement in terms of running speed, trajectory accuracy, and robustness. In order to improve the accuracy of camera trajectory, a loop-closure detection algorithm is proposed based on the sparse direct method. The algorithm directly extracts two features to form a hybrid feature point to improve the robustness of the system. The hybrid feature point is used to track and match key frames, so that the visual odometry can detect closed loops, and then the pose map is used to optimize the positioning accuracy. Experimental results show strong robustness in a complex environment and a balance between speed and accuracy.

Based on the impulsive control method, a robust H_∞ synchronization problem for a class of complex networks with multi-weights is studied, and a novel distributed impulsive controller is designed. By adding the error state feedback item between the node state variable and the synchronous state in the traditional distributed impulsive controller, the robust H_∞ synchronization of the complex networks with multi-weights are guaranteed when they were affected by external interference. Based on the stability theory of Lyapunov, mathematical induction and other relevant knowledge, the sufficient conditions for networks to achieve robust H_∞ synchronization is given in the form of linear matrix inequalities (LMIs). Finally, a numerical simulation verifies the validity of the conclusion.

Let $ A$ be the disk algebra, $ u$ and $ \varphi $ be analytic functins on A, some estimates of weighted composition operators $ u{C_\varphi }$ from Bloch type spaces to the disk algebra are characterized. At the same time, the path connectedness of the space of weighted composition operators is investigated and a conclusion is derived that any two bounded weighted composition operators are pathwise connected.

Te adaptive orthogonal greedy decomposition algorithm in the reproducing kernel $W_2^1[a, b]$-space is studied. The optimal n-term approximation function is adaptively constructed based on the principle of the fastest energy descent, and the convergence of this algorithm is proved theoretically. Finally, an experiment is used to verify that in the reproducing kernel $W_2^1[a, b]$-space, the best n-term numerical original function constructed by the orthogonal greedy principle has a better convergence effect than the best n-term numerical original function constructed with equal division nodes.

Considering the characteristics of location and capacity problems such as multi-variable, multi-constrained, high-dimensional and nonlinear, a method based on Voronoi diagram and improved gravitational search algorithm (IGSA) is proposed. Considering both subjective and objective weights, a multi-objective decision model considering construction costs, operation costs, user charging time costs and distribution network loss costs is established. In order to solve the slow convergence and the inadequate accuracy of standard gravitational search algorithm (GSA), a chaotic map is introduced in the particle initialization and updating phase to enhance the ergodicity of the algorithm. At the same time, a global optimal point to guide speed updating formula is introduced to improve the ability of the algorithm to jump out of the local optimal. Voronoi diagram is used to divide service area of charging station, and a joint solution process of Voronoi diagram and IGSA is proposed. The simulation results prove the feasibility and validity of the model.

Using UV-LED / Cl₂ Advanced Oxidation Process to study the degradation of metoprolol (MTP), one of the β-blocker, the effects of light intensity, chlorine dosage, pH, TBA and other factors on the degradation of MTP are investigated. Studies have shown that UV-LED / Cl₂ can effectively remove MTP by synergy. The MTP degradation during UV-LED / Cl₂ treatment satisfies pseudo-first-order kinetics, and the pseudo-first-order rate constants of MTP (k_obs) increases with the UV-LED light intensity and the pH of the solution. With the increase of the chlorine dose, the k_obs of MTP increases first and then decreases. The k_obs of MTP decreases with the increase of the initial concentration of MTP and TBA, and TBA has a strong inhibitory effect on MTP degradation during UV-LED / Cl₂ treatment.

The implementation of liquid-vapor separation in the phase change process can achieve heat transfer enhancement. Liquid-vapor separation was applied to inverter air conditioning system. The optimal refrigerant charge amount and capillary length of the system were obtained through matching experiments. At the standard operating condition, the obtained EER and COP are 2.57 and 3.68, respectively. The system performance is also examined under different cooling and heating conditions. The annual performance factor (APF) of actual operation in different cities was compared to reflect the actual operating performance of the system. The results showed that the highest APF value was in Tengchong, at 3.67, and the lowest in Shenyang, at 2.81. It shows that the performance of the system varies significantly in different cities. Moreover, the performance at the standard operating condition largely differentiates that at the actual conditions.

The elevation accuracy of AW3D30 SRTM3 v4.1 and ASTER GDEM V3 are analyzed through the area of the mountainous hills in parts of Guangdong. The data of 730 km lines along Guangzhou, Huizhou, Shaoguan and Qingyuan are collected by vehicle dynamic PPP, the WGS84 coordinates of dynamic points calculated by CSRS-PPP positioning service system, geodetic height of dynamic points converted into normal height by gravity field model EIGEN-6C4, and finally, the elevation of three digital elevation models are checked. The results show that the average errors of AW3D30, SRTM3 v4.1 and ASTER GDEM V3 are, respectively, 0.55, 0.17 and 1.59 m, and the root mean square 3.78, 5.84 and 8.88 m. The average error amplitude of AW3D30 in different altitudes is relatively small, within 2.18 m; the average error of SRTM3 v4.1 has a negative correlation with altitude, and the average error gradually changes from positive value to negative value with the increase of altitude; the average error of ASTER GDEM V3 is within (0 m, 2.18 m] altitude range and is about 2 m and -2.28 m in (250 m, 800 m] range. The root mean square error and standard deviation of AW3D30 decrease with the increase of altitude, while SRTM3 v4.1 increases with the increase of altitude. ASTER GDEM V3 has no significant rule, which is better than 7.69 m in (100 m, 250 m], and better than 9.86 m in other regions.

Soil is a kind of geotechnical engineering material with porous media, and the microscopic pore structure determines macroscopic permeability. To investigate the responsibility between permeability characteristics and pore structure, the falling-head permeability tests at various lime contents are carried on lime-treated loess firstly, and then the Scanning Electron Microscope (SEM) is conducted on some typical specimens which use the Image Pro Plus 6.0 (IPP) Image processing software to count the microscopic pore structure parameters of the improved loess at various lime contents, and, by combining the results of the laboratory permeability test, establish the connection between the macroscopic permeability characteristics and the microscopic pore structure. The results show that with the increase of the lime contents, the crystalline cement filling the large and medium pores is increasing, corresponding to the number, area, surface porosity of the large and medium pores and average diameter of the large pores in the microscopic image are gradually reduced, but the number and area of small and micro pores and the fractal dimension of pores are on the rise, and the pore structure tends to be complicated, which reduces the proportion of effective seepage channels. Therefore the permeability of the soil is reducing. It is indicated that microscopic pore structure parameters are closely related to macroscopic physical properties, which can reflect changes in macroscopic physical properties to a certain extent.

The construction of characteristic towns has been spread out in China, which is an important breakthrough in the current urban-rural integration. At present, there are many problems in the prophase planning of characteristic town, such as the lack of objective quantitative analysis and the difficulty of providing more scientific basis and support for the planning and design. Combining the advantages of spatial analysis technology with the needs of planning and design, the application and development of spatial analysis technology in the prophase planning of characteristic town are explored. Taking the planning of Qiantan characteristic town in Jiande City of Zhejiang Province as an example, the geographic information system (GIS) is used for terrain analysis, multi factor land suitability and value analysis, three-dimensional skyline analysis, and the spatial syntax is used for the overall and local traffic network analysis and visual field analysis to establish the scientific basis and technical support for the prophase planning of the characteristic town.