50 results about "Forest classification" patented technology

The invention discloses a forest classification method based on an object-oriented high-resolution remote sensing image. The method is based on the high-resolution remote sensing image, an object-oriented image classification method is used, an orienting remote sensing forest second-level classification system is established, a forest remote sensing classification auxiliary data set and an integrated image are created, key indexes which can distinguish forest types are selected, and a layered step-by-step classification extraction method is provided to be used for establishing information extraction knowledge rules of various forest types. The processes of the method are suitable for middle-small-scale forest resource remote sensing monitoring in a zone, good operability and repeatability are achieved, and efficiency and accuracy of forest remote sensing monitoring in the zone can be effectively improved.

A cityscape image semantic segmentation method based on multi-feature fusion and Boosting decision forest includes the following steps of carrying out super-pixel segmentation on images, carrying out multi-feature extraction, carrying out feature fusion and carrying out training learning and classification recognition. The method effectively integrates 2D features and 3D features and remarkably improves recognition rates of targets. Compared with the prior art, segmentation results are consistent, connectivity is good, edge positioning is accurate, a Boosting decision forest classification mechanism is introduced, and stability of target classification is guaranteed.

The invention discloses a constraint condition-based random forest recommendation algorithm. The algorithm comprises the steps of generating a binary decision tree according to a data set in a given initial training set; randomly selecting features to form a random forest; measuring time efficiency of a decision tree algorithm; measuring efficiency of a random forest algorithm; and improving the random forest algorithm. Compared with other recommendation algorithms, the random forest recommendation algorithm has the advantages that the superiority of the random forest classification algorithm is fully understood, and the random forest classification algorithm is autonomously realized, so that the understanding of a recommendation system is deepened; and finally, key points causing algorithm deficiencies are researched from algorithm construction, and the algorithm is improved, thereby enabling the algorithm to have better efficiency.

The invention provides an emotion classification method based on deep forest and transfer learning. Firstly, a source domain data set and a training target domain data set are selected. Then the deepconvolution neural network is used to train the data set in the source domain, and the trained feature extraction model is obtained and saved. Preprocessing the training target domain data, wherein the preprocessing comprises channel conversion and size clipping; The feature extraction model is used to extract features from the preprocessed training target domain data set to obtain sample features. The sample characteristics are used as the input of the deep forest classification model, and the deep forest classification model is trained. When the training of the depth forest model classification model is completed, the trained depth forest classification model is used to classify the facial emotion images that actually need to be processed, and the classification result of the facial emotion images is obtained. The invention has the advantages that the classification efficiency is improved, the classification cost and the demand for training samples are reduced.

The invention provides a classification method and device for a hyperspectral remote sensing image and storage equipment. According to the classification method for a hyperspectral remote sensing image, dimension reduction processing is performed on a high-dimensional hyperspectral remote sensing image, the training performed by sample data for a deep forest classification model is combined, and then a purpose of accurately and quickly classifying the hyperspectral remote sensing image is achieved through the trained deep forest classification model. The classification device for the hyperspectral remote sensing image and the storage equipment are used for implementing the classification method for the hyperspectral remote sensing image. The beneficial effects of the invention are that thehyperspectral remote sensing image is classified by using a deep forest algorithm, the classification accuracy and the classification speed of the deep forest classification model are improved, and the method provided by the invention involves fewer parameters, is easy to adjust and greatly improves the classification effect and practicability of the hyperspectral remote sensing image.

The invention discloses a method and a system for detecting LDoS attack data flow in a cloud environment. The method comprises the steps of simulating various attacks with network simulation softwareand extracting corresponding network traffic; sampling and classifying the data packets that arrive and lose on the server; extracting the average number of packets, source IP growth rate, wavelet features and congestion participation in a given period of data stream from the sample; analyzing the eigenvalues and establishing the random forest classification model, testing the established random forest model with normal data stream and abnormal data stream, calculating the corresponding information entropy for each feature classification, adjusting the threshold value of eigenvalues accordingto the information entropy, so as to achieve the best classification effect; according to the stochastic forest classification model, designing and deploying the LDoS attack data flow detection systemon the cloud server to detect and process all kinds of LDoS attack data flow in real time. The invention has the advantages of low energy consumption, high detection rate and low false alarm rate, and has high practical value.

The invention discloses a ground target classification method based on a random forest and data rejection, which comprises the following steps: (1) preprocessing a training sample set; (2) extractinga training feature matrix; (3) training a random forest classifier; (4) preprocessing the test sample; (5) extracting test feature vector; (6) calculating the output probability vector of the ground target; (7), judging whether that t sample is rejected or not; (8) if the judgment is rejected, taking the test sample as the echo signal without the fretting characteristic of the target; (9) if the judgment is not rejected, outputting the corresponding class of the maximum value in the probability vector as the ground target classification result of the test sample. The method rejects the judgment of clutter, deceptive interference and non-fretting target signal, improves the classification and recognition rate of ground moving target, and adopts a classifier with parallel processing abilityto improve the real-time performance of the method.

The invention provides a quick CU partitioning and intra-frame decision-making method for H.266 / VVC. The method comprises the steps: firstly, calculating the texture complexity of a current CU throughemploying a standard deviation, and dividing the current CU into a smooth region, a common region or a complex region according to the texture complexity; secondly, encoding the smooth region and thecommon region by using an original prediction mode, and outputting optimal intra-frame mode prediction; utilizing a random forest classifier model to classify the complex region; and finally, performing intra-frame mode prediction on the complex region by using a quick intra-frame prediction mode optimization method based on texture region features, and outputting an optimal intra-frame mode. According to the invention, the intra-frame coding prediction is carried out on the CU through the fast intra-frame prediction mode optimization method based on the random forest classifier and the texture region features, the optimal intra-frame coding mode is obtained, the calculation of more than half of the coding mode can be reduced, the calculation complexity is reduced, and the coding time issaved.

The invention relates to a forest information remote sensing and automatic extracting method based on vegetation index time series data dispersion measures. According to the forest information remote sensing and automatic extracting method, based on vegetation index time series data of each day in a year of each grid pixel element in a research area, according to the overall distribution condition and the distribution condition in different value ranges of the index time series data, an overall dispersion measure index, an intermediate-high dispersion measure index, a growth peak period dispersion measure index and a high value continuity index are designed; based on the principle the vegetation index time series data dispersion of a forest is small, forest classification flow chart is established, forest information is remotely sensed and automatically extracted, and finally a forest distribution map of the research area is obtained. By the adoption of the forest information remote sensing and automatic extracting method, based on the process of fully extracting the changes of vegetation index data dispersion measures of different forest types on the whole within different value ranges and different time periods, multiple dispersion indexes are established and used for forest information remote sensing and automatic extracting, and the forest information remote sensing and automatic extracting method has the advantages that the robustness is good, the classification accuracy is high, the automation degree is high, and the disturbance resistance is high.

The invention discloses a cloud platform decision forest classification method based on discrete weak correlation, which is used for weak correlation of a decision forest classification method of a cloud platform. The method comprises the steps of selecting a random sampling attribute element group according to correlation degree of data attributes, updating the continuous attribute of the probability discretion of the attribute group in a gain manner, and solving a maximum gain attribute; and finally establishing a cloud platform decision forest by virtue of an acquired classification attribute sequence. When a great amount data is processed, the time and space expenditure for establishing the cloud platform decision forest can be reduced, the capability and stability for resisting the data noise can be improved, and the classification prediction speed and the classification quality can be improved.

The invention discloses an isolated forest classification method used for credit risk monitoring and integrated with a deep belief network, and the method comprises the steps: employing the deep belief network to achieve the feature learning of high-dimensional data, and obtaining the optimal low-dimensional representation of an original data set; combined with linear time complexity, an unsupervised learning algorithm-Isolated forest algorithm for anomaly detection; And finally, achieving optimization of important parameters of the isolated Sensen algorithm through a particle swarm optimization algorithm and a simulated annealing algorithm, and forming a final DBN- iForst model for credit risk monitoring, And identifying the loan default.

The invention discloses a method for eliminating city building pixels in a forest classification result based on a PALSAR radar image, relating to the remote sensing image processing technology field. The method for eliminating city building pixels comprises steps of (1) extracting forest information from a PALSAR radar image, (2) finishing optical image pre-processing and calculating a normalization vegetation index NDVI, (3) performing maximum value synthesis on multiple NDVI images to obtain a cloudless NDVI product in a research area, (4) performing re-sampling on the PALSAR forest classification result or the NDVI synthesis product to realize unification of spatial resolution, (5) utilizing the ground samples of the forest and the city building to draw NDVI histograms of two surface features and determine a filtering boundary value T, and (6) performing wave band operation on the PALSA forest result after re-sampling, and filtering the forest pixels with NDVI values being lower than the boundary value T. The invention utilizes the normalization vegetation index NDVI to eliminate or reduce the city building elements of the forest classification result based on the PALSAR, and improves the forest classification result accuracy.

A method and apparatus for cell recognition based on a stochastic forest classification model are disclosed. The method comprises the following steps of: training a random forest classification modelby using an original cell image sample set, testing the random forest classification model to obtain an optimal sample accuracy rate; taking the stochastic forest classification model corresponding tothe optimal sample accuracy rate as a fitness value calculation function, wherein the optimal artificial fish is obtained by artificial fish swarm algorithm, and the initial values of the preset number of trees and the number parameter of the feature subset of the stochastic forest classification model are updated, and the above steps are repeated until the optimal eigenvalue pair is no longer changed. The cells in the image to be detected are classified using a stochastic forest classification model corresponding to the optimal eigenvalue. The present application utilizes artificial fish swarm algorithm for feature selection of a random forest classifier, and solves the problem of feature redundancy in the model and insufficient generalization ability of the whole classifier.

The invention discloses a full-waveform onboard laser radar terrain classification method based on ReliefF feature selection. The method comprises the steps of performing Gaussian decomposition on onboard laser radar full-waveform data, acquiring an echo parameter and optimizing the echo parameter by using an LM algorithm; extracting onboard laser radar data point-cloud features based on differentgeometric neighborhood, and extracting the waveform features of the full-waveform data by utilizing the optimized echo parameter; integrating the point-cloud features in the onboard laser radar datawith the waveform features of the full-waveform data, and evaluating the significance of each feature in the classification by using the ReliefF algorithm; and performing feature selection, analyzinginfluence on different terrain classification effects by multiple feature combinations, and performing terrain classification by utilizing a random forest classifier. The classified optimal feature combination is searched through feature extraction and selection by utilizing the onboard laser radar point-cloud data and the full-waveform data, thereby reaching an aim of realizing high-precision terrain multi-classification.

The invention discloses a multi-feature optimization and fusion method for crop planting structure extraction, and the method comprises the steps: collecting a time sequence satellite remote sensing data set which is not greater than the monthly scale, and completing the pre-obtaining of sample data in a research region; describing spectral and texture characteristics of various crops; calculatingexpressions of different samples on spectral information, vegetation indexes, texture characteristic quantities and the like, counting mean values and variances of the characteristic quantities, andcalculating distinguishable capabilities of the different samples on the characteristic quantities; establishing a multi-feature optimization formula, and determining feature quantities participatingin classification and proportions of the feature quantities in the classification process by utilizing the formula; constructing a new image; and performing fine identification on the crop type of theresearch area by utilizing a random forest classifier, generating a space-time distribution thematic map of the crops, and verifying the precision. According to the method, the problem that the timecomplexity and the computer running speed are increased due to the fact that screening of the classification characteristic quantity is ignored in a traditional remote sensing information extraction method is solved.

The embodiment of the invention discloses a machine pump fault diagnosis method and device based on random forest classification. The method comprises the steps: acquiring Z-axis vibration signal components in a machine pump fault state and a normal state; preprocessing the Z-axis vibration component to obtain the frequency domain signal; then training the frequency domain signal by adopting a random forest method, and establishing a relationship between the frequency domain signal and a fault state to obtain a fault diagnosis model; and inputting a to-be-diagnosed pump frequency domain signal into the fault diagnosis model, and judging the fault state of the pump. According to the method, by using random forest classification, calculation is convenient, the diagnosis speed is high, and the real-time performance and accuracy of fault diagnosis are improved; therefore, the method reduces the cost of fault diagnosis, and has a good practical value.

The invention discloses a man-made forest spatial pattern recognition method based on time sequence classification and spatial analysis, and belongs to the technical field of forest classification andrecognition. According to the method, on the basis of multi-source remote sensing data collection and preprocessing, field measured data are combined, backward scattering features for identifying land cover categories are extracted from a multi-year PALSAR image, and vegetation index NDVI cumulative maximum values of different land cover categories are extracted from a multi-temporal Landsat image for phenological feature identification; by integrating forest and non-forest products obtained by the former and forest and non-forest category discrimination thresholds of the latter, a single forest resource category identification model is constructed and extended to consecutive years, and spatial distribution identification of the artificial forest is realized by combining decision discrimination, knowledge criteria and spatial overlay analysis. Compared with a traditional field investigation shape identification and area statistics mode, the method shows clear space-time significance characteristics.

The invention discloses a no-reference image quality evaluation method based on deep forest classification. The method comprises the following steps: step 1, image classification; step 2, extracting color quality characteristics of the image; step 3, extracting texture quality characteristics of the image; step 4, simulating the difference of different people on image quality cognition by utilizing the difference of decision tree extraction features in the deep forest classification model, and constructing the deep forest classification model to classify the image quality, including a multi-granularity scanning forest and a cascade forest; step 5, training the deep forest classification model based on the image quality features and the category labels thereof to obtain the probability thatthe test image belongs to different categories, i.e., statistical information of subjective evaluation results of different people on the image quality; step 6, setting a quality anchor, and fully considering the difference in the subjective evaluation process in combination with the probability that the image belongs to different categories to obtain a final image quality score. According to thenon-reference image quality evaluation method, the difference of different people for image quality cognition is simulated by using the deep forest, so that an image quality evaluation result is given. The method has important theoretical significance and practical value.

The invention provides an electric power system disturbance classification method, which can capture robust feature expression of lost data in disturbance data based on a feature extraction method of a stacked denoising auto-encoder, and on the basis, a random forest classifier is used to realize identification of electric power system disturbance. The method can quickly and accurately classify the PMU disturbance data, still has very high recognition accuracy for the PMU disturbance data containing lost data, and is good in anti-noise performance. Compared with an existing disturbance classification method, PMU disturbance data containing lost data can be classified quickly and accurately, and real-time monitoring of dynamic behaviors of a power system is achieved.

Methods and systems which provide super-resolution synthesis based on weighted results from a random forest classifier are described. Embodiments apply a trained random forest classifier to low-resolution patches generated from the low-resolution input image to classify the low-resolution input patches. As each low-resolution patch is fed into the random forest classifier, each decision tree in the random forest classifier “votes” for a particular class for each of the low-resolution patches. Each class is associated with a projection matrix. The projection matrices output by the decision trees are combined by a weighted average to calculate an overall projection matrix corresponding to the random forest classifier output, which is used to calculate a high-resolution patch for each low-resolution patch. The high-resolution patches are combined to generate a synthesized high-resolution image corresponding to the low-resolution input image.

The invention provides a parallel depth forest classification method based on information theory improvement. Firstly, the algorithm designs a hybrid dimension reduction strategy based on the information theory, a data set after dimension reduction is obtained, and redundancy and irrelevant feature numbers are effectively reduced; secondly, an improved multi-granularity scanning strategy is provided for scanning samples, it is guaranteed that all features appear in a data subset at the same frequency after scanning, and the influence of multi-granularity scanning imbalance is avoided; and finally, in combination with a MapReduce framework, the parallel training is carried out on a random forest in each layer of cascade structure of the deep forest model. Meanwhile, a sample weighting strategy is proposed, and a sample with a relatively poor evaluation result is selected to enter the next layer of training according to the evaluation of the random forest in cascade on the sample, so that the number of samples in the layer is reduced, and the parallel efficiency of the algorithm is improved. The method is simple in principle and easy to implement, the operation efficiency and the clustering accuracy are remarkably improved, and the method can also provide great help in biology, medicine and astrogeography.

The invention discloses a wetland vegetation feature optimization and fusion method based on JM Relief F. The method comprises the following steps: collecting an unmanned aerial vehicle high-resolution remote sensing image in an experimental area, and meanwhile, obtaining field sample verification data; describing spectral information, texture features and spatial geometric features of various crops; calculating the expressions of different vegetations in spectral information, texture features and spatial geometric features, and counting the mean value and variance of each feature variable; establishing a JMRelief F multi-feature optimization formula, and determining the weights of the feature variables participating in classification and the separable degree of each feature variable by using the formula; and using a random forest classification algorithm to carry out fine identification on wetland vegetation in a research area, and carrying out precision verification through sample data collected in an experiment area. The method has the characteristics of wide identification range, high efficiency, low cost, short period, high precision and the like. The method can be used in the fields of protection and supervision of wetland vegetation, and can effectively improve the artificial recognition efficiency and precision.

The invention discloses a product cost evaluation method and system based on industrial big data. The method comprises the following steps: collecting industrial big data, and extracting product characteristic data, product processing process data and product processing cost data from the industrial big data; inputting the product characteristic data and the product processing process data into arandom forest classification model for training to generate a processing scheme recommendation model; inputting the product processing process data and the product processing cost data into a CART decision tree for training to generate a processing cost prediction model; and performing product cost prediction by using the processing scheme recommendation model and the processing cost prediction model. Product machining efficiency can be improved, and machining cost can be reduced.

The invention discloses a fruit tree variety identification method based on a visible near infrared spectrum. The invention relates to the fruit tree variety identification method based on the fruit tree leaf visible near infrared spectrum, which is mainly composed of a convolution noise reduction auto-encoder (CDAE) and a random forest (RF), wherein the convolution noise reduction auto-encoder is mainly used for carrying out feature extraction on fruit tree leaf visible near infrared spectrum data; and the random forest classifier is responsible for classifying the features extracted by the convolution noise reduction auto-encoder so as to identify different varieties of fruit trees. According to the fruit tree variety identification method, feature values are extracted by using the convolution noise reduction auto-encoder, the method has the advantages of high classification accuracy, strong noise immunity, good feature extraction capability, omission of a data preprocessing step and no need of spectrum preprocessing, the leaf spectrum is analyzed by using the method, the performance of a random forest algorithm is improved by using the convolution noise reduction auto-encoder, and compared with a traditional random forest algorithm susceptible to noise interference, the method has great progress in the aspect of robustness; and a novel rapid identification method is provided for apple tree variety identification.

The invention discloses a signal random forest classification method, system and device based on decision tree accuracy and correlation measurement, and belongs to the field of signal classification and recognition. The objective of the invention is to solve the problem of low classification accuracy of a single decision tree classifier in a traditional random forest classifier. The method comprises the following steps: firstly, establishing decision trees, verifying each decision tree by using three groups of reserved data sets, calculating the accuracy rate of the i-th decision tree, and sorting all the decision trees in a descending order according to the classification accuracy rate; for the determined data set, adopting a vector inner product method to calculate and store an inner product numerical value between the decision trees, reserving the decision trees of which the vector inner products are smaller than or equal to an inner product threshold value, and otherwise, marking the decision trees with low classification accuracy in each pair of decision trees of which the vector inner products are calculated as deletable; sequentially deleting the decision trees marked as deletable according to the classification accuracy from low to high until the number of the remaining decision trees is N; and performing voting by adopting the final classifier to determine a final classification result. The method, system and device are mainly used for signal classification and identification.

The present invention discloses a method for removing a high biomass sugar cane in a PALSAR forest classification result, relating to the technical field of remote sensing image processing. The method comprises the steps of (1) extracting forest information in PALSAR radar images by using a decision tree classification or supervision classification method, (2) selecting multiple optical remote sensing images which are close to a PALSAR radar image acquisition time and are not in a forest leaf-fall season, completing pretreatment, and calculating a moisture index LSWI, (3) carrying out minimum value integration on multiple LSWI images, and obtaining an LSWI product without cloud in a research area, (4) carrying out resampling on a PALSAR forest classification result or the LSWI product, realizing the unification of spatial resolution, (5) carrying out band operation on the resampled PALSAR forest result, and filtering the forest image element whose LSWI value is lower than zero. The forest classification accuracy is improved.

The invention provides an early warning method and a system for detecting the state of ML standby nodes, and the method comprises the following steps: S1, collecting the physical information and nodestate of each node in a cluster, and generating a random forest classification model in combination with a random forest classification algorithm; S2, acquiring standby nodes in the cluster through database retrieval, and collecting physical information and actual node states of the standby nodes; S3, inputting the physical information of each standby node into the random forest classification model to obtain a prediction node state of each standby node; and S4, comparing the predicted node state and the actual node state of each standby node, and carrying out early warning on the standby nodes. According to the method, early warning of the ML high-availability standby node is realized, so that the standby node is ensured to be in an available state all the time, the success rate of switching between the main node and the standby node is improved, and the safety and the reliability of ML use of a user are ensured.

The invention discloses a no-reference image quality evaluation method based on deep forest classification, comprising: step 1, image classification; step 2, extracting the color quality feature of the image; step 3, extracting the texture quality feature of the image; step 4, using In the deep forest classification model, the decision tree extracts different features, simulates the difference in perception of image quality by different people, and constructs a deep forest classification model to classify image quality, including multi-granularity scanning forest and cascading forest; step 5, based on image quality Features and their category labels, train the deep forest classification model, and obtain the probability that the test image belongs to different categories, that is, the statistical information of the subjective evaluation results of the image quality by different people; step 6, set the quality anchor, and combine the images belonging to different categories The probability of taking into account the differences in the subjective evaluation process to obtain the final image quality score; the no-reference image quality evaluation method described in the present invention uses a deep forest to simulate the difference in image quality cognition of different people, thereby giving an image The quality evaluation results have important theoretical significance and practical value.