With rapid development of big data, mining valuable knowledge may face the impact of high-dimensional, large and dynamic data. The emergence of these complex data streams results in a decline in the classification effect. In order to further analyze the research status and challenges of data stream ensemble classification, the ensemble classification of complex data streams is reviewed for further research. For the first time, it focuses on the core ideas and performance of the algorithm from the perspective of complex data stream and domain data stream; among them, complex data stream mainly introduces concept drift, imbalance, multi-label data stream, etc. Then, text, graph, sensor are introduced, and the application of ensemble learning in domain data streamare sumarizes. Furthermore, data stream evaluation method is introduced from the aspects of verification technology and evaluation indicators. Finally, the future research direction is given, including challenges of the ensemble and classification of uncertain data stream, challenges of the ensemble and classification of multiple data stream, challenges of the delayed data stream and the ensemble classification, and challenges of data stream classification and evaluation methods.

To clarify the research hotspots, directions and trends in the field of soil organic carbon (SOC) at home and abroad are explored by using bibliometric method. The CNKI and WoS databases related to karst SOC research literature in the last 30 years were retrieved, and the co-occurrence analysis of the country, annual publications and journals were carried out with CiteSpace software, as well as highlighting word analysis and cluster analysis of keywords. The literature volume in this field has been increasing year by year in the past 30 years and is in a rapid growth stage. China has the largest number of publications, and the journals Environmental Earth Sciences and China Karst ranking first in the number of publications. The main international research hotspots include“global climate change & soil respiration”, “agroecological management & soil quality”, and “forest ecosystem & soil carbon sequestration”, while the main domestic research hotspots include “karst geomorphic environment & physicochemical properties”, “land management type & vegetation restoration”, and“soil microorganisms & environmental benefits”. In future research, soil microbial properties should be linked closely with karst ecosystem functions, and applied research on hot trends such as stone desertification ecological restoration, agro-ecological health and global climate change in karst should be carried out.

To solve the problem of poor category prediction effect of direct integration of simple classifiers to recognize traffic sign databases, a traffic sign recognition algorithm based on Convolutional Neural Network and Bagging ensemble learning is proposed, using crawler and image enhancement technology to expand the data set of traffic signs, and the features of traffic sign images is extracted by Convolutional Neural Network. For the characteristics of the logo image, the image data are down-sampled by using the maximum pooling layer, and the overall network structure is simplified by using a shallower network depth. On the basis of Convolutional Neural Network feature extraction, the soft voting mechanism is used to integrate Multiple Logistic, K nearest Neighbor, and SVM individual learners to obtain a better prediction result and achieve more accurate traffic sign recognition. Experimental results show that the recognition accuracy of the algorithm on the TSRD traffic sign recognition database test set has reached 93.00%, which is 11.99% higher than the recognition accuracy of the original Network model, and is higher than the recognition accuracy achieved by VGG16 and ResNet50 migration learning having a faster convergence speed.

In order to solve the problems of large output voltage ripple in the bridgeless Boost power factor correction (PFC) and output voltage fluctuations caused by sudden changes in output load, this paper designs a ripple suppression unit based on the bridgeless Boost PFC topology to reduce the DC output Ripple, and to increase the load current feedforward control to optimize the single-cycle control mode. Through theoretical and principle analysis, it can be known that the ripple suppression unit can reduce the DC output ripple, and the load current feedforward control can quickly collect the output current change caused by the load mutation and perform signal conditioning in time to quickly stabilize the output voltage, thereby increasing the dynamic stability of the system sex. Through simulation and comparative analysis, the ripple coefficient is reduced from 2.5% to 0.25% after using the suppression unit. After the load current feedforward control is increased, the stability of theoutput voltage can be restored in a very short time regardless of the sudden increase or decrease of the output load.And at the same time the effect of power factor correction keeps anchangeable. The simulation result verifies the correctness of the theoretical analysis.

To solve the problems of harmonic selectivity (HS) and harmonic foldback (HFB) in traditional N-path filters, this paper proposes a differential N-path filter based on sampling calculation. Through this method, the proposed filter can not only suppress the even harmonics and the third and the fifth harmonics, but also solve the problem of harmonic foldback without increasing the input reference clock frequency. The simulation results show that the filter has a harmonic suppression of 49 dB and 67 dB at the 3rd and 5th harmonics, respectively. The circuit center frequency f_s can be adjusted from 0.3-1.0 GHz, and the total power consumption is 25.2 mW. Input return loss S₁₁<-10 dB, and noise factor (NF) is3.1-5.8 dB.

Aiming at the disadvantages of low voltage gain of traditional converters, high voltage stress of switching devices, and large output voltage ripple.An improved quadratic Buck-Boost converter is proposed by adding a switched capacitor unit and an LC filter in the rear stage of the traditional quadratic Buck-Boost converter. Firstly, the working principle of the new converter in continuous conduction mode (CCM) is analyzed. Then, the new converter is compared with other quadratic converters from different anglesand the results show that the newconverter has good performance characteristics.Finally, a comparison is carried out between theexperiment of the new converter and the experiment of the original converter, and the results show and that the output voltage gain of the new converter is (1+D)/D timesof that of the original one. In the step-up mode with the output voltage of 37.4 V, the output voltage ripple before improvement is approximately 3 times of that after improvement. In the step-down mode with the output voltage of 9 V, the output voltage ripple before improvement is approximately 12 times of that after improvement. At the same time, the voltage stress of the switching device is significantly reduced under the same output voltage. Both of the expermental and theoretical analysis verify that the improved quadratic Buck-Boost converter is feasible.

The research of category text generation is a task to enable machines to generate human-understandable text, and give the text specific attributes. The CNN network has been used to extract the features of the text in most of the existing text generation work based on Generative Adversarial Networks that lacks attention to the global semantics. In addition, the attention mechanism is simply introduced to the generator, which can not effectively eliminate the noise in the decoding process. For this reason, this paper proposes a joint modeling method of local semantics and global semantics. The global semantic information extracted by the LSTM and the local semantic information extracted by the CNN are fused, so that the attention on the semantic information is enhanced during the generation. Moreover, the attention on attention is introduced in the generator to further filter irrelevant information during the sequence generation. Compared with the benchmark models, the method proposed in this paper achieves at least 0.01 and 0.004 improvement in BLEU values on two public real datasets (Movie Review and Amazon Review), demonstrating the effectiveness of the proposed method.

Most of the image captioning models are individuality-agnostic, which cannot generate an individual description according to the user’s intention and language style. To address the above problem, a personalized image captioning model is established in this paper by using fine-grained scene control graph and the style control factors to represent user’s intention and style of speaking, respectively. Firstly, construct a scene control graph, including the object, object attributes and the relationship objects in the scene, which can control the object, object attributes and the relationship between object. Secondly, a multi-graph convolutional neural network is used to encode the context information of the scene, and graph flow attention is employed to control the focus of the description. Then, add the style control module when generating sentences, that is, use keyword search to generate user profile according to user’s gender, age, education level and other information. Finally, the style generator extracts the corresponding style pattern according to the user profile, and the language decoder outputs a personalized image caption. The experimental results on MSCOCO dataset and FlickrStyle dataset show that the proposed method can generate personalized and diverse image caption sentences.

To solve the problems of the large amount of network parameters, low accuracy, and difficulty in deploying the mobile terminal model of the current deep learning model for tomato leaf diseased identification, a tomato leaf disease identification method based on SK-EfficientNet is proposed. The lightweight model EfficientNet is used as the benchmark model, and the selective convolution kernel mechanism SKNet is used to replace the SENet in the EfficientNet core module MBConv, so that the convolution kernel can adaptively chooses the size of the receptive field according to the multi-scale information of the input features. This method improves the ability to extract image features while making more effective use of parameters.Through multiple sets of comparative experiments, the results show that the training accuracy of the improved model has been further improved, and the model parameter is 3.83 MiB.The average accuracy rate on the PlantVillage data set is 99.64%, and the recognition accuracy of SK-EfficientNet-B2 is the highest; in natural scenes, the average accuracy rate is 3.81 percentage point higher than the original model.The results show that the improved model can effectively increase the accuracy of tomato leaf disease recognition in natural scenes, which provide a reference for the deployment of tomato leaf disease recognition models on mobile terminals.

For the parameter estimation of the Ⅲ-conditioned adjustment problem, reasonable equality prior information among parameters helps to improve the accuracy of the model solution. The joint calculation is carried out under the sample information and the prior information of the equation. Based on the morbid least squares adjustment criterion, a new biased estimation algorithm, principal component Liu estimation, is constructed through the principal component estimation and Liu estimation. The parameter solution of theprincipal component Liu estimation with equality constrained least squares is derived, and the formula of correction factor is derived by using the principle of mean square error minimization. The effectiveness and reliability of the proposed method are verified by an example, which can be applied to the problem of solving Ⅲ-conditioned least squares parameters with equality constraints.

The computation for the Hilbert transform of a given signal over a finite interval is an important problem in signal processing. Although the Hilbert spline transform (HST) method with order three demonstrates high order computational accuracy and the speed of O(nlog n), there is a disadvantage of this method: the grid of the B-spline knots must be different from the sample points to avoid singularities in the computation. To solve this problem, the spline function with order four is used to implement the Hilbert spline transform. To prove the effectiveness of the proposed method,several function of the Hilbert transform are calculated by using the cubic spline. Numerical results show that the proposed method has excellent performance in both computational speed and computational accuracy.

Based on the singular value decomposition and spectral decomposition of matrices, the necessary and sufficient conditions for the existence of solutions to rank constraint least square symmetric semidefinite solutions and itsoptimal approximation solution of the matrix equation AX=B are established and, if the solutions exist, the general expression of the solutions are proposed. The computational procedures of the optimal approximation solution, and the numerical examplesshowing the correctness of the theoretical results are given.

A connected graph is said to be C_m-critical if every complete subgraph with order no more than m of G is contained in a smallest separating set. Mader shows that C₃-critical graph is 6-connected. Pastor shows that every component of G₆,induced by the set of vertices of degree 6 of a minimally C₃-critical graph, has a cycle. By using some insight on properties of fragment, it is shown that every vertex of minimally C₃-critical graph adjacent to at least two vertices of degree 6, which implies the result of Pastor. Further, it is shown that everyC₄-critical connected graph is 7-connected.

1,3-Dibromopropane was reacted with alcohol, carboxylic acid and amine respectively to obtain a series of allyl compounds. The one-pot approach was used in this study. Firstly, estradiol as a raw material reacted with 1,3-dibromopropane under the action of alkali then the best conditions for this allylation reaction were determined. Under the best conditions, estradiol was replaced with other alcohols, carboxylic acids and amines for the reaction. The experimental results showed that the reaction of estradiol and 1,3-dibromopropane under the action of NaH alkali produced 3,17β-diallyloxyestradiol-1,3,5(10)-triene instead of 3-bromopropyl. The reaction of 1,3-dibromopropane with other alcohols, carboxylic acids and amines got higher yield of allyl ethers, allyl esters and allylamines under relatively mild conditions.

In order to study the chemical constituents and biological activities from the bark of Toona ciliata var. pubescens, in this paper, silica gel column chromatography, Sephadex LH-20, reversed-phase ODS C₁₈ and MCI, semi preparative HPLC and other chromatographic techniques were used to isolate and purify the natural compounds; AGAR dilution method was used to screen the bactericidal activity of plant pathogens, and the antitumor activity was screened by MTT method. Twenty compounds were isolated from the ethyl acetate extract of 95% ethanol extract of the bark of Toona ciliata var. pubescens, and were identified to be toonacilin (1), odoratone (2), bourjotinolone A (3), dyvariabilin H (4), piscidinol A (5), piscidinol B (6), 3-episapelin A (7), 8(14), 15-isopimaradiene-2α,3α,18-triol (8), labd-13-ene-8α,15-diol (9), nidorellol (10), catechin (11), 1-linolenoylglycerol (12), tianshic acid (13), (-)-neoechinulin A (14), aurantiamide acetate (15), dibutylphthalate (16), 4,7-dimethoxy-5-methylcoumarin (17), scopoletin (18), (22E)-5a,8a-epidioxyergosta-6,22-dien-3b-ol (19), β-sitosterol (20), respectively, based on physicochemical properties and spectral analysis techniques.Except forcompounds 4, 11, 16~18, 20 ,the remaining compoundswere isolated from this plant for the first time.The results of antibacterial activity showed the compound 1 had potential inhibitory effect on Aternaria solani, Botrytis cinerea and Phytophthora capsici at a concentration of 50 mg/L, whilecompound 4 on Botrytis cinerea was 76.5% at a concentration of 50 mg/L . Compound 2 showed strong inhibition against human breast cancer cell line MDA-MB-231 (IC₅₀=12.9 μmol/L ), andcompound5 showed strong inhibition against HEL cellwith IC₅₀ value of 8.4 μmol/L. Compound 3 had strong inhibitory effects on human breast cancer cell line MDA-MB-231, human rhabdomyoma cell line A-673 and human erythrocytic leukemia cell line HEL, with the IC₅₀ values from 8.2 to 11.2 μmol/L.

To construct a stable knocking down of Slfn2 expression cell line in Lewis lung carcinoma (LLC) using CRISPR/Cas9 system, a guide RNA (sgRNA) sequence targeting the Slfn2 gene was designed by using the CHOPCHOP website software. After ligation of the sgRNA to the lenti-CRISPR-v2 plasmid, the T293 cells were co-transfected with pLenti CRISPR v2-Slfn2, pMD2.G and psPAX2 to generate Lentivirus. The packaged lentivirus was used to infect LLC cells, and Slfn2^-/-knockout LLC cells were obtained by screening and Monoclone culture. The results showed that the two SgRNAs were successfully inserted into the vector respectively. Slfn2 gene in LLC was successfully mutated by lentivirus packaged with three plasmids. Two Slfn2 knockout LLC cell lines were obtained, one with 8 bp deletion plus frameshift mutation and the other with 180 bp deletion mutation. The mouse lung cancer cells knocked out by Slfn2 gene were successfully constructed by CRISPR/Cas9 technology.

Chinese crocodile lizard Shinisaurus crocodilurus is critically endangered with small population size. Although captive population is increasing, this species is still at risk of inbreeding due to their small population size. Therefore, it’s necessary to understand the genetic structure and relationship of this species to avoid inbreeding. Based on 9 pairs of microsatellite markers, the genetic structure of 112 S. crocodilurus in Luokeng Nature Reserve artificial breeding population was analyzed. The results indicated that: (1) Microsatellite loci detected a total of 106 alleles, with the average of 12 alleles for each loci. (2) Hardy-weinberg exact test are within the range of 0.314-0.998, the balance test coefficient is greater than 0.05. (3) The range of observed heterozygosity (Ho) is 0.018-0.929, with an average of 0.629, and the expected heterozygosity (He) ranges from 0.018 to 0.851, with an average of 0.622. (4) The individual Neighbor-joining tree of S. crocodilurus based on the Nei’s genetic distance showed that they clustered into two clades with three and five lineages in close genetic relationships. (5) Bottleneck effect detection results showed that under the TPM model, the captive population of S. crocodilurus from Luokeng Nature Reserve has experienced bottleneck effect in the recent history. These results indicated that the S. crocodilurus population in Luokeng Nature Reserve had an extremely low genetic diversity. It is recommended to consider genetical relationship in future artificial breeding to avoid inbreeding and keeping the genetic diversity.

Animals need various nutrients to satisfy their nutritional requirements, and their feeding strategies are affected by nutritional requirements. In order to study the relationship between nutrient contents and food choice of Franois’ langur Trachypithecus francoisi in limestone forest, from September, 2016 to August, 2017, the feeding behavior of langur in Guangxi West Daming Mountain Autonomous Region Nature Reserve was studied by instantaneous scanning sampling method, and it was divided into staple food and non-staple food according to the proportion of feeding time. Nutritional contents included water, crude protein, acid detergent fiber (ADF), neutral detergent fiber (NDF) and minerals (Zn, Ca, K, Mg, Cu, Fe, Mn). The results indicated that there are 23 species of food, including 9 staple food and 14 non-staple food, with feeding proportion ranged from 0.02% to 6.81%. In young leaves, the contents of K, Mg, Mn, NDF and ADF were significantly different between staple food species and unstaple food species, but there were no significant difference among other nutrient elements in young leaves. In mature leaves, the contents of Cu, Mg and Mn of staple food species were higher than that of non-staple food species, but there was no significant difference among other nutrient elements in mature leaves. The generalized linear model (GLM) showed that Mg, NDF and water content had significant effects on the Franois’ langur feeding proportion. The Model averaging further indicated that the contents of the two nutritional components, Mg, and water, jointly determined the food choice of Franois’ langur.

Seed removal behavior of rodents is of great significance to the regeneration and diffusion of plant population. The seeds of Kmeria septentrionalis and Cyclobalanopsis glauca with different sizes were placed in four karst microhabitats (stone cavern, stone groove, stone surface, and soil surface) from September to October, 2019.The differences of seed removal rate in different microhabitats were compared to explore the effects of karst microhabitat and seed size on rodent removal behavior. The results showed that there was a significant difference in the removal rate between the seeds of K. septentrionalis and C. glauca (Z=-21.902, P<0.001), and rodents preferred to remove the seeds with smaller volume (93.56%±0.99% vs 7.41%±0.65%). The removal rates of K. septentrionalis were significantly different from those of C. glauca in all same microhabitats (P<0.001), and the average seed removal rate in the microhabitats of stone cave (53.41%±3.60%), stone groove (51.31%±3.66%) and stone surface (49.19%±3.59%) were higher than that of soil surface microhabitat (45.37%±3.59%). Microhabitat and seed size are important factors that affect the seed removal of rodents, and the results show that the removal behavior of rodents to plant seeds is the result of the interaction of karst microhabitat and seed size.

Under the condition of long-term management, the soil obstacle problem of eucalyptus plantation in the lower tropical region frequently occured. The identification and evaluation of soil fertility obstacle factors in main eucalyptus producing areas based on soil types, and the development of regional management and fertilization strategies for forest rights owners can provide a theoretical basis for the improvement of soil fertility, which is of great significance to ensure the high quality and sustainable development of Guangxi plantation industry. In this paper, a total of 298 typical soil samples of red soil, yellow soil and red soil in the main planting areas of eucalyptus plantations in northern Guangxi in the past three years were collected and analyzed. Cluster analysis combined with principal component analysis was used to screen the main obstacle factors from 15 soil fertility factors. The obstacle degree model was introduced to quantitatively analyze the obstacle factors of different soil types. The results showed that the main obstacle factors were different in different soil types. In red soil, there were four severe obstacle factors, which were pH, total phosphorus, available manganese and available boron, among them, pH barrier was the highest (28.20%), and soil total nitrogen was the mild disorder factor. In yellow soil, pH, total nitrogen, available manganese and available zinc were the severe obstacle factors, among them, pH barrier was the highest (36.53%), and total potassium was the mild disorder factor. In the red soil, there were three severe obstacle factors, total potassium, available magnesium and available zinc, among them, available magnesium barrier was the highest (40.72%), and the mild obstacle factor was organic matter and available manganese. Due to the widespread application of formula fertilizer in short-cycle plantation management in Guangxi, the available N, P and K nutrients of the three main soil types in this region are relatively sufficient, and the soil problems are concentrated in acidification, low phosphorus, partial trace elements loss and other common problems. However, due to the different site conditions in different regions, it is suggested that the forest right owners should develop business strategies according to the local conditions and soil types in order to reduce the occurrence of soil obstacles.

In order to adapt to the special environment and water quality in the high cold and anoxic area, the anaerobic MBBR process was selected to treat the low temperature rural wastewater, and the startup characteristics of three kinds of fillers were explored by investigating the morphology of biofilm on the fillers, the treatment effect on actual wastewater, the biofilm proliferation rate and the respiratory rate per gram of Levapor sponge filler, K3 filler and polyurethane sponge filler. The results showed that the biofilm appeared inside the K3 packing and outside the porous packing at first,then became black anaerobic biofilm with the consumption of dissolved oxygen in the reactor.The removal efficiency of ammonia nitrogen and total phosphorus in anaerobic reactor was low,and the removal rates were -15%-8% and -10%-15% at the last stage, respectively. The porous sponge fillers had higher COD removal rate, faster biofilm proliferation speed, but had shorter service life and higher endogenous specific respiration rate, while the removal rate of COD by K3 filler increased steadily, and K3 filler had slower biofilm formation speed because of its faster biofilm renewal speed, but had lower endogenous specific respiration rate, and longer service life under mechanical stirring, so it is more suitable for the biological carrier of anaerobic MBBR reactor.

Green space is one of the basic elements of a city, which plays a certain role in alleviating urban environmental problems. The study on the spatial-temporal evolution and driving force of the overall green space in Fujian Province is helpful to the construction of ecological civilization demonstration area in Fujian Province. Based on the land cover data, socio-economic and natural data of Fujian Province in 2000, 2010 and 2020, the dynamic attitude calculation, transfer matrix and PLSR model were used to analyze the evolution law and driving mechanism of green space at different scales in Fujian Province. The results show that: ①from 2000 to 2020, the overall green space area of Fujian Province show a downward trend, while the non-green space increase. The area transfer largely between cultivated land, forest, grassland and construction land, and the transferred locations are mainly located in the coastal areas of Fujian Province. ②The changing trend of the overall green space in nine regions is basically the same as that in Fujian Province, that is, the green space decreases and the non-green space increases. There are differences between the evolution of green space types in different regions: the areas with reduced green space are mostly located in the coastal areas of Fujian Province. ③Regression model results show that the national economy and population growth in the province have an important impact on all kinds of green spaces(VIP＞1), while the forces of other influencing factors are different to some extent. Social and economic factors have an important influence on the green space in coastal areas. Natural factors have different influences on different types of green spaces, and natural factors have different influences on green spaces in different regions. The analysis reveals that economic development, the proportion of industrial structure and the growth of population promote green space to turn to non-green space, and the differences of social and economic development in different regions also lead to spatial differences in the evolution of green space.