420results about How to "Valid reservation" patented technology

The invention relates to a Computerized Tomography (CT) image uniformization metal artifact correction method based on four-order total-variation shunting, wherein solving is carried out on the basis of a convexity splitting method to achieve metal artifact correction. The CT image uniformization metal artifact correction method includes the specific steps of collecting data, reconstructing images, dividing metal zones, calculating priori images, obtaining re-projection of the metal zones and the priori images, uniformizing projection data, performing four-order total-variation equation correction, reversely uniformizing the corrected data, reconstructing images again and restoring the image metal zones. The CT image uniformization metal artifact correction method can effectively remove metal artifacts, well maintains structural information of metals and the periphery of the metals, and limits secondary artifact from appearing to the maximum extent and the like.

The invention discloses a co-channel full-duplex system based on MPPSK modulation. After the system isolates receiving and sending signals at the radio-frequency head normally, cancellation conducted on leakage self-interference signals is finished on a digital baseband; the cancellation method is conducted after conducting shock filter on MPPSK receiving and sending aliasing signals, extracting and multiplying by coherent carrier, band-pass filtering and digitizing, the result of initial channels of all self-interference channels between the output end of a modulator and a receiver ADC is estimated before formal communication, self-interference offset signals are rebuilt by the adoption of the estimated result, further residual errors are filtered out with the combination of methods of coding rate filtering and double matched filtering of a method of shock filtering-multiplying by the coherent carrier-band-pass filtering-coherent demodulation, and MPPSK receiving signals are demodulated reliably. The co-channel full-duplex system based on MPPSK modulation is free of radio-frequency self-interference cancellation, has low requirements for the dynamic range and sampling rate of the ADC, has high isolation degree for the self-interference signals, and is simple in structure, low in complexity, high in spectrum efficiency and good in demodulation performance.

The invention provides an image salient target detection method combined with deep learning. The method is based on an improved RFCN deep convolution neural network of cross-level feature fusion, anda network model comprises two parts of basic feature extraction and cross-level feature fusion. The method comprises: firstly, using an improved deep convolution network model to extract features of an input image, and using a cross-level fusion framework for feature fusion, to generate a high-level semantic feature preliminary saliency map; then, fusing the preliminary saliency map with image bottom-layer features to perform saliency propagation and obtain structure information; finally, using a conditional random field (CRF) to optimize a saliency propagation result to obtain a final saliency map. In a PR curve graph obtain by the method, F value and MAE effect are better than those obtained by other nine algorithms. The method can improve integrity of salient target detection, and has characteristics of less background noise and high algorithm robustness.

The invention discloses a hyperspectral image classification method based on deep learning, and belongs to the technical field of remote sensing image processing. The method comprises the steps that 1, dimension reduction treatment on a hyperspectral image is achieved by obtaining a data sample, conducting layer-by-layer training on an autoencoder network and further adjusting an initial weight value obtained through pre-training by adopting a BP algorithm; a data cube in each pixel neighbourhood in the hyperspectral image is taken as input of a convolutional neural network, a ground object type corresponding to a pixel serves as expected output of the convolutional neural network, the convolutional neural network is trained, the trained convolutional neural network acts on the whole hyperspectral image, and a final high-precision classification result is obtained. According to the method, the defects that in a traditional hyperspectral image classification problem, details are discarded in the dimension reduction process, space information is lost in the classification process, and the classification precision is low are overcome, the good classification precision is achieved, and the method is suitable for classification of various hyperspectral images.

The invention belongs to the field of image processing methods, and discloses a three-dimensional roof reconstruction method based on LiDAR data and ortho images. The method comprises the following steps of: (1) LiDAR point roof facets segmentation based on a triangular cluster; (2) roof ridge extraction based on LiDAR data nad ortho images; (3) three-dimensional roof model reconstruction. In the method, for meeting the requirement on accurate reconstruction of the three-dimensional roof model, the LiDAR data and the high-resolution images are integrated, the complementary advantages of the elevation characteristic of the LiDAR data and the high-resolution characteristic of the images are comprehensively used, 'roof facets segmentation, ridge extracton and three-dimensional roof model reconstruction' is taken as a principal line, a LiDAR point roof facets segmentation algorithm based on the triangular cluster and a roof ridge extraction algorithm based on the LiDAR data and the ortho images are realized, and a new method for three-dimensional roof model reconstruction is formed. Tests show that the method of the invention has high automation degree, higher accuracy and integrity and high location precision in modeling, and meets the needs of actual application.

The invention discloses an adaptive compressed sensing-based non-local reconstruction method for a natural image. The problems of serious reconstructed image information loss and the like in the prior art are mainly solved. The method is implemented by the steps of: (1) dividing an image into N 32*32 sub-blocks, obtaining a basic sensing matrix Phi' according to a basic sampling rate b and a sensing matrix Phi, and sampling a signal by utilizing Phi' to obtain a basic observation vector; (2) estimating a standard deviation sequence {d1, d2, ..., and dN} of the image according to the basic observation vector; (3) adaptively allocating a sampling rate ai for each sub-block according to the standard deviation sequence {d1, d2, ..., and dN}, and constructing an adaptive sensing matrix, and sampling the signal by utilizing the adaptive sensing matrix to obtain an adaptive observation vector; (4) forming an observation vector of each sub-block by using the basic observation vector and the adaptive observation vector; (5) obtaining an initial solution x0 of the image according to the observation vector; and (6) performing iteration by using x0, and reconstructing the original image until consistency with a finishing condition is achieved to obtain a reconstructed image x'. The method has the advantages of high image reconstruction quality, clear principle and operational simplicity, and is applied to the sampling and reconstruction of the natural image.

The invention discloses an open domain video natural language description method based on multi-modal feature fusion. According to the method, a deep convolutional neural network model is adopted forextracting the RGB image features and the grayscale light stream picture features, video spatio-temporal information and audio information are added, then a multi-modal feature system is formed, whenthe C3D feature is extracted, the coverage rate among the continuous frame blocks input into the three-dimensional convolutional neural network model is dynamically regulated, the limitation problem of the size of the training data is solved, meanwhile, robustness is available for the video length capable of being processed, the audio information makes up the deficiencies in the visual sense, andfinally, fusion is carried out aiming at the multi-modal features. For the method provided by the invention, a data standardization method is adopted for standardizing the modal feature values withina certain range, and thus the problem of differences of the feature values is solved; the individual modal feature dimension is reduced by adopting the PCA method, 99% of the important information iseffectively reserved, the problem of training failure caused by the excessively large dimension is solved, the accuracy of the generated open domain video description sentences is effectively improved, and the method has high robustness for the scenes, figures and events.

The invention discloses a multi-spectral imaging system and a multi-spectral imaging method based on double cameras. The system comprises a system supporting module, a light source module, an optical signal acquiring module and a computer module, wherein the system supporting module is used for supporting and connecting various components; the light source module is used for providing infrared light and visible light; the optical signal acquiring module is used for acquiring fluorescent light and visible light images; and the computer module is used for controlling the optical signal acquiring module to acquire images, processing the acquired images and displaying the processed images. Light which passes through a lens is divided into two parts by an optical beam splitter prism, the two parts of light are simultaneously acquired by the two charged coupled device (CCD) cameras in real time, problems in the prior art are solved effectively, the technological monopoly of foreign companies in China is broken, the barrier of multi-spectral video imaging research is lowered, the selectable space for optical molecular image probes is expanded, and the optical molecular image research and application range is also expanded.

The invention discloses a video-image-resolution-based vision adaptive method. The method comprises the following specific steps: (1) inputting an original video image and extracting image attention; (2) setting a clipping ratio, and finding an optimal clipping window area, remained by a video, of each frame with greatest total energy; and (3) adjusting the clipping ratio, finding an optical clipping ratio to obtain an optimal clipping window area, clipping the video image, and scaling the video image to preset target resolution. By the method, under the preset target resolution of the video image, main information of the video image can be remained; when the target resolution is much smaller than the original resolution, a main object in the original video image does not become very small; and when the aspect ratio of the target resolution is not the same as that of the original resolution, the length-to-width ratio of the object does not change, and the video image is clipped and scaled according to the optimal clipping window area with the optimal clipping and scaling ratio to provide a good vision effect for a viewer.

A hemp seed protein powder is prepared from hemp seeds through removing impurities, classifying, quick heating, quick cooling, mechanical peeling, cold squeezing, separating protein flakes from oil, extracting and separating protein 4-6 times, vacuum evaporating to remove solvant, pulverizing and screening.

The present invention discloses technology of enzymolyzing waste bamboo shoot coat to prepared functional diet fiber. Waste bamboo shoot coat, after being washed, dried, crushed and defatted, is added into pH 4.5-5.5 phosphate buffering solution in the concentration of 15-20 ml / g with cellulolytic enzyme in enzyme activity of 5-10 U / ml and enzyme concentration of 0.5-1 % to enzymolyze at 100-120 rpm and 50-60 deg.c for 1-2 hr, before decolorizing and vacuum drying to obtain functional diet fiber. The simple technological process can utilize the fiber resource in waste bamboo shoot coat effectively, and the functional diet fiber product high soluble diet fiber content.

The invention belongs to the technical field of digital image processing, provides a TV (total variation) image noise removal method based on noise priori constraint and can improve the noise detection accuracy and well protect image structural information. The technical scheme is that the TV image noise removal method based on the noise priori constraint comprises the following steps: step 1) inputting a noise-containing image I; step 2) enabling f to be an original sharp image; step 3) performing impulse noise influence on an image u in transmission and storage processes, and representing an image polluted by mixed noise with g; step 4) estimating noise positions with an ROAD (rank-ordered absolute difference) statistical method; step 5) constructing a TV-ROAD iteration noise removal model; step 6) solving an equation shown in the specification according to a noise removed image f obtained in the previous step to obtain a two-value matrix vector b. The method is mainly applied to digital image processing.

The present invention provides a method and an apparatus for simplifying the point cloud of the apple leaf. The method comprises: using a bounding box method to perform fast K-nearest neighbor search,establishing a kd-tree space storage structure of the point cloud, setting different thresholds to identify the point cloud boundary of the leaf and extracting the point cloud boundary of the leaf; and by calculating the normal vector, the curvature, and the like of the feature parameters of the point, calculating neighborhood point position information, distinguishing feature points and non-feature points, and further carrying out simplification processing on the non-feature points. According to the technical scheme of the present invention, boundary point clouds and non-boundary point clouds can be quickly and conveniently obtained, and a simplification result can be further obtained; during the process, different K values and multiple thresholds can be set according to requirements, the obtained point result accuracy is relatively high, the calculation process is convenient and the calculation method is reasonable, the obtained point result accuracy is suitable for automatic programming, the waste of computer resources is effectively reduced, and the operating efficiency can be improved to some extent.

The invention discloses a method for eliminating tailing light lines of a frame transfer type CCD (charge coupled device) sensor in a star image, which is used for solving the technical problem that the SNR (signal to noise ratio) of a weak and small target is reduced due to increment of noise points in an existing method for detecting tailing light lines. The technical scheme is as follows: a two-dimensional WT (wavelet transform) method is adopted to decompose an original image into low-frequency sub-images and vertical, horizontal and diagonal high-frequency sub-images, and CCD tailing is vertical or horizontal light lines, so that the low-frequency sub-images and the vertical or horizontal sub-images comprise the light lines, thus only eliminating the light lines of the sub-images and effectively reserving other detailed information; a detection method based on array grey and singular values is used in the process of detection, thus the complexity is low; and the influences on excessive light stars are considered sufficiently in the process of elimination, thus the estimation of grey intensity of the light lines is more accurate. Because an interpolation method is not used, the SNR of the weak and small target is not reduced; and on the contrary, the SNR of targets on the tailing light lines is increased by over 5dB on average.

The invention discloses an image super-resolution reconstruction method based on a multi-band deep convolutional neural network, comprising the following steps: selecting training samples and test samples; inputting training images to the network in batches for feature extraction, and combining long-term and short-term memories to construct a multi-band learning structure, then performing featurerefinement, feature mapping and up-sampling reconstruction; and obtaining network parameters according to the model obtained through training to complete the image reconstruction. The invention enhances high-frequency information reconstruction by introducing multi-band learning, and uses the memory migration operation to make the network have long-term and short-term memories at the same time, which speeds up the image reconstruction speed, enhances image edge and texture detail reconstruction, and obtains image super-resolution reconstruction results of better quality. The invention has strong super-resolution capability, and the reconstructed image is closer to the real image.

The invention discloses a neural network construction, training and recognition method, a system, and a storage medium. Due to discontinuity of a neuron internal state variable and an error function,an input layer pulse sequence is constructed by using a convolution kernel function of a time pulse. In the hidden layer, the membrane voltage of synaptic neurons is solved by utilizing an electric leakage-integration-excitation neuron model. When the membrane voltage exceeds a predefined threshold value, one pulse is excited to be transmitted to the next layer. In a network output layer, trainingerrors are calculated through a difference value between a desired pulse sequence and a real pulse sequence. After that, a Widrow-Hoff (WH) learning rule is simulated to carry out gradient error reduction and back propagation. The neural network model has strong capability of processing complex time series data (such as target recognition and voice recognition in a video), and is helpful for realizing the application of brain-like calculation in practice.

The invention provides a preparation method of a mesoporous silicon dioxide hollow sphere, relates to a silicon dioxide hollow sphere, and provides a preparation method of a mesoporous silicon dioxide hollow sphere. The method comprises the steps of: adding a silicon dioxide sphere into water, and ultrasonically processing to obtain silicon dioxide sphere dispersion solution; adding a cationic surface active agent into the obtained silicon dioxide sphere dispersion solution, adding an alkali source, and etching in a stirring way; and collecting, cleaning and drying a precipitate after etching to obtain white powder, and removing the cationic surface active agent out of the white powder to obtain the mesoporous silicon dioxide hollow sphere. The silicon dioxide sphere is taken as a template, and the mesoporous silicon dioxide hollow sphere is prepared by means of carrying out alkali etching under the effect of the cationic surface active agent. The preparation method has the advantages of being strong in operability, low in cost, simple in reaction device, mild in preparation processing condition, cleaning and pollution-free in reaction processe, and high in reaction efficiency and the like. Compared with other synthesis method, the preparation method is good in industrial application prospect.

The invention, which belongs to the technical field of computer vision images, discloses a local-information-based infrared weak target detection method in a complex background. According to the invention, a filtering structure is constructed by using one point v0 in a to-be-detected image as the center, wherein the filtering structure includes a protection frame and a filtering window; a plurality of directions that use the point v0 as the center and point to the filtering window are found out, wherein the plurality of directions are evenly distributed; weighted gray values from the protection frame to the filtering window in the multiple directions are calculated respectively and one value closest to the point v0 gray value among obtained results is used as a background estimation valueof the point v0; the filtering window slides in the to-be-detected image to obtain background estimation value of all points and a background estimation image is formed; the background estimation image and the to-be-detected infrared image are differentiated to obtain a difference image; and then target segmentation and extraction are carried out on the difference image to obtain a target detection result. With the method provided by the invention, clutter interferences of the edge information and the edge intersection part in the complex background can be suppressed effectively; adaptation tothe bright and dark targets is improved; and the target detection effect is improved effectively.

The invention discloses an SAR image speckle suppression method based on second generation curvilinear wave transformation, which mainly overcomes the defect of scratch effect and point target loss brought by a curvilinear wave to the SAR image speckle suppression. The SAR image speckle suppression method comprises the following steps: performing the second generation curvilinear wave transformation to a selected test image and partitioning the selected test image into 5 layers of subbands; keeping coefficients of the first layer unchangeable and zero-setting coefficients of the fifth layer; respectively evaluating parameter vectors of hybrid Gaussian models from the second layer to the fourth layer by an EM method; marking the coefficients from the second layer to the fourth layer; reconstructing the image, detecting the edge of the reconstructed image, and performing the average filtering to the uniform area of the reconstructed image to obtain the filtered image; and performing the nonlinear anisotropy dispersion iteration to a difference image obtained by the original image and the filtered image to obtain a speckle suppressed image. The invention has the advantages of keeping clean edge of the image, removing the scratch effect and remaining the point target characteristic information of the image, and can be used for preprocessing scene analysis and image understanding in the SAR image.

The present invention discloses a preparation method for a leisure bean product, ie., multi-layer bean rolls. The purpose of the present invention is to overcome disadvantages of single variety, rough mouthfeel, poor toughness and the like of the existing leisure bean product. The main process steps of the method of the present invention comprise: selecting materials, soaking and cleaning, refining, centrifugating and filtering, carrying out continuous slurry boiling, carrying out slurry curdling, beating into fine slurry, carrying out first molding, carrying out alkalization seasoning, carrying out second molding, and carrying out secondary seasoning. According to the present invention, the traditional bean product processing technology and the modern food production technology are effectively combined, and the production process is optimized, such that problems of alkalization molding, seasoning, retention of soybean nutrients and the like are solved, and the technical level of the products is effectively improved; the product of the present invention has characteristics of delicate tissue, high elasticity, moderate hardness, unique flavor, rich nutrition, convenient eating, health and safety, easy carrying and storing, and the like, and is suitable for various consumer groups.

The present invention relates to a method, an apparatus and a system for network resource reservation and a mobile terminal therefor. Provided are a method, apparatus and system for minimizing unnecessary network resource reservation by predicting an expected direction of a terminal. The method for network resource reservation based on a moving speed of a mobile terminal includes: receiving a request for network resource reservation from the mobile terminal; calculating an expected moving direction of the mobile terminal, based on a speed-based moving probability model represented by a probability density function for a movable direction dependent on the moving speed of the mobile terminal, a moving speed of the mobile terminal, and position information of the mobile terminal; and performing network resource reservation for wireless cells included in the calculated expected moving direction. Thus, unnecessary network resource reservation can be minimized by predicting the moving direction of the terminal.

The invention relates to a method for extracting a natural flavor with dimethyl ether, which comprises the following steps: making leaves, fruits, flowers and seeds of aromatic plants into a pulp and freeze-drying the pulp, or cleaning and drying stems, barks and roots of aromatic plants, and pulverizing and granulating; sending the freeze-dried material or pulverized and granulated material into an extraction tank, extracting to obtain an extracting solution by using dimethyl ether as a flavor extracting agent, filtering the extracting solution to remove granular impurities, pumping to an evaporation tank, and separating the extracting solution under reduced pressure with a compressor; and carrying out rough separation on the residues after the separation under reduced pressure, carrying out fine separation with a molecular distillation instrument, and distilling by 2-5 stages to obtain the aromatic essential oil. The invention overcomes the defect of low efficiency in the traditional method and the defect of high cost in the supercritical method; the used dimethyl ether extracting agent is odorless and smellless, so the invention has the advantages of no environment pollution, low price and high extraction efficiency; and the obtained plant essential oil is clear and transparent, contains little wax, and overcomes the defects of high water content, high wax content and high oxidative deterioration tendency in the traditional plant essential oil.

The edible fruit vinegar preparing process includes setting various kinds of pure fruit juice in liquid state deep fermenting tank, adding proper amount of liquor or edible alcohol and honey and inoculating acetobacter to ferment at certain temperature for certain time to prepare raw vinegar; ultrafiltering, adding mineral water, essence, etc and blending to form vinegar product of 3.5-5% acidity. The edible fruit vinegar contains rich amino acids, minerals and vitamins, and may be used as seasoning and beverage. It can promote body's metabolism and replenish various kinds of trace elements and vitamins.

The invention provides animage salt-and-pepper noise removal method based on a mean value in an iteration switch. By setting a simple noise detection operator, a point-by-point self-adaptive filtering window is conveniently constructed, and noise points are rapidly judged. Due to the fact that pixel gray level values in small areas of an image have high relevance, by means of the small-size 3*3 pixel filtering window, detail information of the image can be effectively reserved, and meanwhile the false detection problem caused by the simple noise detection operator becomes very small. High-concentration salt-and-pepper noise is removed through a layer-by-layer switch filtering method instead of a large-size filtering window, and the noise can be eliminated step by step. Under the high-concentration noise environment, the gray level values of the noise points are replaced with mid values in grey level values of non-noise points, in all directions, in a long distance, mean value filtering is conducted on filtered pixels, and deviation is further reduced. By means of the method, the aim of reserving existing detail information of the image while effectively removing noise is achieved.

The invention provides a processing method of nutritional frozen Hibiscus esculentus slice. The method comprises the steps of material selection, color protection, sterilizing (ozone combined with microwave), freezing (cooling, and ultrasonic-assisted freezing), thawing, preparing, screening, slicing, quick-freezing, packaging, sealing and delivering. The method provided by the invention adopts ultrasonic-assisted freezing to solve tissue structure disintegration problem in single freezing process, effectively promotes nucleation under simultaneous mass-transfer action and heat-transfer action of ultrasound, avoids damage of functional ingredients in Hibiscus esculentus, and avoids nutritional ingredients loss by adopting two-step operation of freezing and sterilizing. Under joint action of ozone and microwave in sterilization process, the method has thorough sterilization and obvious effect; in combination with sterilization with 75% ethanol, bacterial count can be controlled at 500-300 / g, sterilization time is shortened by above 30%, and the product is safe, pollution-free and residue-free.

The invention relates to an image noise eliminating method for preserving high frequency information, and is characterized in that: comprising the two following steps that: (a) linear filtering processing is done to an inputting image with random noise and a filtering function: (b) conjugate gradient optimization restoration operation is done by taking the filtering result of the frontal step as a degenerated function, wherein a first step is to fully eliminate the random noise in the image, a second step is to restore the high frequency information such as a losing image edge, details and so on of the first step. Therefore, the image noise eliminating method of the invention is capable of fully eliminating the image random noise on the premise of preserving the image high frequency information.

The invention discloses a making method of semi-fermented fruit wine. The fruit wine is produced in a base wine immersion and fermentation mode; fruits are immersed to dissolve nutrition components and pigments in the fruits in order to effectively reserve the fruit aroma, reach oxidation prevention and color protection effects and realize sterilization and bacteriostasis, and fermentation is carried out to increase the mellowness of the fruit wine an make the wine have plump and soft mouthfeel, so the semi-fermented fruit wine has mellow and plump soft mouthfeel and well reserves the color and the rich aroma of the original fruits, and the defects of two traditional fruit wine processing technologies are effectively overcome, thereby the fruit wine has high nutrition values and good mouthfeel, the health efficacy of the fruit wine is in favor of improving people's health, and the fruit wine is easily accepted by consumers. The raw materials of the fruit wine have wide selection range, resources are fully utilized, and seasonal fruits can be selected to make the fruit wine, so the cost is low; and the method has the advantages of simplicity, easiness in operation and short production period, and the fruit wine is a low-alcohol degree alcohol and beverage integrated health fruit wine, and has wide application prospect.

The invention provides an image obtaining method and an image obtaining device. The method and the device can realize multi-focus shooting and processing on shooting subjects with different depths of field, and images with a high clear degree are obtained. The method provided by the invention comprises the following steps that: a plurality of regions to be focused are recorded according to received selection instructions of the regions to be focused; each region to be focused is subjected to focal plane contrast ratio calculation; then, automatic focusing is carried out according to the focal plane contrast ratio, and a plurality of focal lengths to be focused are obtained; shooting is performed by referring to the plurality of focal lengths to be focused, and a plurality of shot images are obtained; each shot image is divided into a plurality of sub images in the same meshing mode; the plurality of sub images corresponding to each mesh region are subjected to sub image contrast ratio calculation; for the plurality of sub images corresponding to the same mesh region, the sub image with the highest sub image contrast ratio is selected to be used as a fusion sub image corresponding to the mesh region; and the fusion sub images corresponding to all mesh regions are spliced, and a fusion result image is obtained.

The invention provides a method and system for extracting a significant target region of an image based on iterative sparse representation. Firstly, superpixel segmentation is performed on an originalimage by using an SLIC segmentation method with a plurality of groups of different numbers of pixel element parameters to generate a group of split images with different sizes of superpixel regions;for segmentation results of each scale, taking classical visual attention detection results as the initial saliency map constrained foreground and foreground sample region selection, calculating reconstruction residual of each superpixel region as a significant factor by the sparse representation process, optimizing significant test results under the single dimension in combination with recursiveiterative operations, and finally obtaining final significant targets and test results through multi-scale saliency map fusion. According to the method and system for extracting the significant targetregion of the image based on iterative sparse representation, the shortcomings of inconsistency in the single-target saliency evaluation, the difficulty in detection of an image edge significant target, and the incomplete extraction of multiple significant targets in a conventional method are effectively overcome.

The present invention provides a gas concentration prediction method based on an adaptive modularization neural network, and relates to the mine gas concentration detection technology field. The method comprises: collecting the gas concentration data, storing the gas concentration data in a gas concentration database, performing adaptive noise removing processing of the gas concentration data in the database to take the gas concentration data after the adaptive noise removing processing as chaotic time series, establishing the training sample set of an adaptive modularization neural network, constructing an adaptive modularization neural network soft measurement prediction model, and predicating the gas concentration by employing the constructed predication model according to the newly obtained gas concentration data and the historical data in the gas concentration database. The gas concentration prediction method based on the adaptive modularization neural network has a significant noise removing effect, can retain the useful information in the gas concentration time sequence while effectively removing the noise and can construct the soft measurement predication model of the adaptive modularization neural network, and the input information is integrally processed through a plurality of different submodels so as to improve the learning precision and the generalization performance of the predication model and improve the robustness of the prediction model.