135results about How to "Reduce preprocessing" patented technology

The invention provides a lane line detection method based on a full-convolution network and relates to the field of traffic information detection. According to the method, probability operation is performed on an output feature map of the full-convolution lane line detection network, the probability that a lane line appears in each region in an input picture is obtained, a prediction probability threshold is set, and extraction and detection of lane lines are realized. Through the method, detection of straight lane lines and curved lane lines can be realized simultaneously; a lane line detection loss function is used to train the full-convolution lane line detection network, so that the detection effect of the lane lines is improved; the convolutional neural network learns abstract features of the lane lines from a lane line classification dataset instead of simply extracting external features of the lane lines; detection of a newly input image can be realized just by storing a lane line detection network model, so that storage space is saved, and the method is suitable for vehicle-mounted embedded equipment; and the small and shallow full-convolution lane line detection network isadopted to perform detection acceleration, so that detection speed is high.

The invention discloses a method for detecting a ship in an aerial image of an unmanned aerial vehicle based on deep learning. The method comprises the steps that firstly, aerial images of unmanned aerial vehicles including military ships and civil ships are collected and marked, and a ship database is obtained; Then, the obtained ship database is sent to a deep learning network for training untilthe network converges; And finally, a ship target in the aerial image of the unmanned aerial vehicle is detected by using the trained deep learning network and the weight file, and outputting a detection result. The method has better accuracy and robustness, effectively solves the problem of small target detection in the aerial image of the unmanned aerial vehicle, also solves the problems of environmental interference, illumination influence and low accuracy of ship detection in a traditional image processing algorithm, and can be suitable for ship detection in different scenes.

The invention pertains to the technical field of treatment of waste water, and particularly relates to a method of hydrolysis-catalytic iron-aerobic coupling for treating poisonous and harmful hard-degradation waste water, which comprises the following steps: waste water after primary treatment directly enters an up flow type hydrolysis system 1 and stays for 4.0-10.0 hours; the effluent enters an electrolytic cell in the catalytic iron; the mass ratio of iron scraps and copper scraps in the electrolytic cell of the catalytic iron is 1:1 to 20:1, the bulk specific weight thereof is 0.1 to 0.3, the contact time thereof in a filling area is 0.5 to 2.0 hours, and the reflux ratio of the effluent refluxing to the up flow type hydrolysis system is 20 to 100 percent; the effluent enters an aerobic biological treatment system and stays for 6.0 to 12.0 hours; then the effluent enters a solid-liquid separation system with a reflux ratio of sludge being 10 to 50 percent; and the effluent of the solid-liquid separation system refluxes to the electrolytic cell of the catalytic iron with the reflux ratio being 20 to 200 percent. Under the premise of fully playing the function of different treatment units and maintaining the treatment effect and respective advantages, the treatment method of waste water realizes the coupling synergetic action of the three units, strengthens the impact of the system to water quality and water quantity load, reduces a plurality of steps of pretreatment, effectively improves the removal rate of ammonia nitrogen, and can be flexibly adjusted.

The invention discloses a device for separating a light-colored C9 fraction of cracking, which comprises a sleeve tower, a condenser, an overhead product accumulator tank and a tower bottom reboiler, wherein the inside of a rectifying tower is sleeved with an internal rectifying cylinder body, so the whole rectifying tower is divided into four parts, namely a public rectifying area, a public stripping area, the internal rectifying cylinder body and the external rectifying tower body, and the inside of the rectifying tower is provided with a tower plate or a filler; the part above the internal rectifying cylinder body is the public rectifying area, the top part of the public rectifying area is provided with an outlet for C6 to C8 fractions, and a raw material inlet or at least one outlet for the light-colored C9 product of the cracking are arranged in the internal rectifying cylinder body; and the part below the internal rectifying cylinder body is the public stripping area, and the bottom part of the public stripping area is provided with a heavy component outlet. The raw material is introduced from the middle upper part of the external rectifying tower body and is separated primarily in the external rectifying tower body, the separation of light components from medium components is completed in the upper part of the internal rectifying cylinder body, and the separation of heavy components from the medium components is completed in the lower part of the internal rectifying cylinder body. Compared with the traditional separation method, the method can save about 30 percent of energy and improve product yield.

The present invention discloses a production process of hot-rolled sheet for cold roll drawing operation. Said production process includes the following steps: smelting molten iron by using top and bottom double-blast converter, controlling oxygen supplying intensity, binary alkalinity and converting time, adopting double-slag operation, refining molten steel by using LF furnace, adding refining slag, in the refining process adding reaction inhibitor, adding alloy elements of boron, titanium and aluminium, making molten steel undergo the process of microalloying treatment, then making continuously-casting and continuously-rolling operation so as to obtain the invented hot-rolled sheet.

The invention discloses a method for super-rapidly preparing metal-organic framework films through using an external electric field method. The method comprises the following steps: immersing a substrate into a solution of metal salts and organic ligands, applying an external electric field, and directly synthesizing dense metal-organic framework films in one step through a direct current method. The method provided by the invention is used for growing the metal-organic framework films on a porous conductive substrate through the external electric field method, is simple and efficient, and has ultrahigh synthesizing rate and good universality; the processes of activation, pretreatment and the like are eliminated; the operation process is simplified; the production efficiency is greatly improved; meanwhile, the method has the advantages of low cost and low energy consumption; the low-cost and easily available substrate material is combined with the simple and efficient external electric field method; the simplicity and operability of equipment are provided for large-scale production; the large-scale industrial production is facilitated.

In order to solve the problem of lane recognition in current vehicle automatic driving, a neural network model is planned to recognize a lane as a final solution. Compared with a typical computer vision technology in which a gradient threshold value, a color threshold value and other manual programming methods are used for processing pictures, the deep learning model can achieve a better effect inthe aspect of extracting key characteristic values of the images. In particular, a full convolutional neural network model, namely a neural network without any full connection layer is constructed; important specific pixel groups are found by using a filter, lane feature information in an image can be extracted very effectively, and finally output information can be displayed in a lane area recognized by an original image in a highlight manner, so that the purpose of lane recognition is achieved.

The invention provides a high-resolution remote sensing image fusion evaluation method based on diffusion tensor guidance. The method includes the steps that a multispectral image and a panchromatic image which are subjected to image registration are segmented, an anisotropic model is adopted to carry out up-sampling under diffusion tensor guidance, a remote sensing image fusion algorithm is selected, GPU parallel computing acceleration is adopted, and remote-sensing image fusion evaluation index correlation analysis is carried out. According to the method, the action intensity of a diffusion tensor model on image pixels can be regulated with parameters in the up-sampling process, and thus edge protection and noise removal can be balanced according to actual demands; up-sampling is not influenced by the imaging quality such as the brightness and the contrast ratio of an original image, strict correspondence of overlapped images is achieved, and the phenomenon of false structures of pixels near the edge is avoided; besides, a generalized model is proposed to generalize the fusion strategy, and the quality of an obtained fused image is comprehensively evaluated through correlation analysis; the performance of existing equipment is fully exerted, and multithread processing is carried out by utilizing the computing advantage of high strength of a GPU.

The coke oven gas adsorption reinforced catalytic reaction process for producing hydrogen includes the following steps: 1. heating fresh catalyst in a reactor to decompose and introducing hydrogen to reduce the catalyst; 2. introducing coke oven gas into the reactor to produce catalytic hydrogen producing reaction; 3. regenerating catalyst when it is needed; and 4. repeating the steps 2 and 3. The present invention converts CO, methane and other lower carbon hydrocarbons into hydrogen, raise the utilization rate of methane and other impurity in coke oven gas. Through reaction conversion, the hydrogen concentration in coke oven gas is raised from 50-60 % to over 90 %.

The invention discloses a text categorization method based on probability word selection and a supervision subject model. The method includes the following steps that (1) punctuation marks in a training text are removed, word frequency information and category information are collected, and a word list and a category list are formed; (2) a subject proportion vector, a subject word matrix, a subject word distinguishing degree matrix and a regression coefficient matrix are initialized; (3) the subject proportion vector, the subject word matrix, the subject word distinguishing degree matrix and the regression coefficient matrix are updated according to the word list of the training text and category iteration of the word list; (4) for a testing text, the word frequency information is collected, and categorization is conducted by using the subject proportion vector, the subject word matrix, the subject word distinguishing degree matrix and the regression coefficient matrix. By means of the method, a complex pre-processing process during text categorization can be reduced to the maximum degree, and the testing text can be categorized more accurately. By means of the method, the distinguishing degree of words in a subject can be excavated, and the significance of the words in the text can be displayed visually.

The invention relates to the technical field of image segmentation, and provides a segmentation method of an iris region in an iris image based on a Mask R-CNN neural network which comprises the stepsof S100, S100, establishing an improved Mask R-CNN neural network for iris segmentation S200, before training the neural network, marking a sample iris image set for training; S300, respectively inputting the labeled sample iris images into a neural network, and training the neural network until convergence; S400, inputting the iris image that needs to be segmented by the iris region into the improved Mask R-CNN neural network that completes the training, and acquiring the double circle boundary information and the binary mask map of the iris region; and S500, completing iris segmentation according to the double-circle boundary information of the iris area obtained in the step S400 and the binary mask map. The invention realizes automatic learning to obtain a normalized iris image and aniris region binary mask by using the improved Mask R-CNN neural network in iris segmentation and localization. Accurate iris area double-circle boundary information can also be obtained, and subsequent iris recognition operation is facilitated.

The invention discloses an electroencephalogram spatial-temporal feature learning and emotion classification method based on a hybrid neural network. The method comprises the steps of collecting electroencephalogram signals of multiple channels; extracting PSD (power spectral density) features from the electroencephalogram signals of the multiple channels; converting the features into a two-dimensional mesh matrix sequence; dividing the sequence into a plurality of fragments Pj; establishing a CASC_CNN_LSTM model and a CASC_CNN_CNN model, jointly extracting deep spatial features and temporal features of the electroencephalogram signals from each fragment Pj through the CASC_CNN_LSTM model, and inputting the deep spatial features and temporal features extracted by the CASC_CNN_LSTM model into a softmax layer corresponding to the CASC_CNN_LSTM model to perform emotion category prediction; jointly extracting deeper spatial features of the electroencephalogram signals from each fragment Pjthrough the CASC_CNN_CNN model; and inputting the deeper spatial features extracted by the CASC_CNN_CNN model into a softmax layer corresponding to the CASC_CNN_CNN model to perform emotion categoryprediction. According to the invention, the emotion classification is more accurate.

The invention relates to a high gradient magnet separation device for the separation or purification of magnetic or magnetically labeled biological material, comprising a magnet (7), a separating column (1), and a ferromagnetic matrix (2) that can be arranged in an interior chamber of the separating column, and a storage container (11) containing a buffer solution for the equilibration of the separating column (1) and/or for the suspension of the biological material, wherein during operation a magnetic field created by the magnet (7) creates a high gradient magnetic field in the matrix (2) and buffer solution can flow from the storage container (11) through the separating column (1). For this purpose, the buffer solution comprises a base solution and macromolecules that can saturate nonspecific binding sites of the matrix (2).

A method for amplifying image data is disclosed. Aiming at the problem of imbalance of target detection samples in depth learning, the over-fitting phenomenon caused by the problem is avoided by increasing the amount of data of fewer samples (the category with fewer samples), and the generalization ability of the model after training is improved, at the same time, the amount of image data acquisition and the workload of pre-processing can be reduced. The method comprises the following steps of: determining a category of a desired amplified image; extracting a category mark frame in the image through a labeling software; determining a target area and a noise interference area; changing pixel values of the noise interference area; and generating a new sample. By using this method, new samples can be generated from the existing samples, and the disturbance of noise features to the model can be reduced while the target features are highlighted. Moreover, the generated new samples do not need to be pre-processed again, so the detection model can be trained directly.

The invention discloses a method for improving lignocellulose enzymolysis and saccharification efficiency, which comprises the following steps: mixing sheared lignocellulose material, ethanol, sulfuric acid and water, and heating in a high-temperature and high-pressure resisting sealed reactor to carry out pretreatment reaction to obtain a mixture; neutralizing the obtained mixture by using an alkali, separating solid from liquid, and distilling the liquid, wherein the component distilled out is ethanol, and the liquid left in a distiller is a solution rich in pentose; drying the solid, adding cellulase and water, and adjusting the pH value, and performing enzymolysis, wherein the liquid re-separated from the solid is the solution rich in the glucose. The method has the advantages that the pretreatment temperature is lower than the pretreatment temperature of the common diluted acid, the content of hemicellulose and lignin in the solid obtained through pretreatment is low, the conversion rate of glucose is high due to enzymolysis, the used ethanol can be recycled, and the fermentation inhibitor in sugar liquid is less pretreated than the common diluted acid.

The invention provides micro-nano lignocellulose, a preparation method and applications thereof. The micro-nano lignocelluloses contain lignin structures; lignin is bonded with cellulose through hydrogen bonds and chemical bonds. The preparation method comprises following steps: a lignin containing cellulose raw material is dispersed in a hot urea aqueous solution so as to obtain a cellulose raw material dispersion liquid; mechanical pre-treatment is adopted for peeling and grinding so as to obtain a pre-treatment product; and the pre-treatment product is subjected to high pressure homogenizing using a high pressure homogenizer so as to obtain a micro-nano lignocelluloses dispersion liquid. According to the preparation method, removing of the lignin in the lignin containing cellulose raw material is not needed, the hot urea aqueous solution is combined with mechanical pre-treatment and high pressure homogenizing so as to prepare the product; the diameter of the obtained micro-nano lignocelluloses ranges from 5 to 180nm; the length/diameter ratio is 200 or larger; the lignin content ranges from 10 to 35wt%; micro-nano cellulose performance is improved; and the application prospect is promising.

The invention relates to a convolutional neural network-based physical characteristic recognition and classification-based clothing recommendation system. The system comprises an automatic customer information acquisition module, a clothing style recommendation module and a clothing information management module, wherein the automatic customer information acquisition module is used for obtaining photos of customers through a man-machine interaction manner, extracting physical features of the customers in the extracted photos by adoption of a CNN-SVM multi-classifier algorithm-based image recognition technology, and taking the physical features as input information of the clothing style recommendation module; the clothing style recommendation module is used for carrying out personalized clothing style recommendation by adoption of an expert system technology according to the physical features of the customers, and generating clothing recommendation lists; and the clothing information management module is used for completing order payment management on the basis of the generated clothing recommendation lists. According to the system, clothing recommendation can be completed accordingto the physical features of the customers.

The present invention is light aggregate pre-treating process and apparatus. The light aggregate is first pressurized and humidified in a sealed pressure container as the light aggregate pre-treating apparatus in the pressure of 5-8 MPa for 30-120 min to reach water absorbing rate of 60-80 %; and then shell forming treated through mixing with cement for 90-180 min, with the cement adding amount being 50-100 kg for 1 cu m of aggregate without free water, to form watertight shell in the surface of the aggregate. The present invention has high aggregate pre-treating speed, high efficiency, and capacity keeping free water amount and maintaining high pumpability of the concrete.

The invention discloses a method for producing hydrogen by directly reacting and absorbing the enhanced methane steam with sulfur-containing methane and by reforming. The sulfur-containing methane and the steam are introduced into a reactor and react to generate the hydrogen under the action of a composite catalyst. The composite catalyst comprises the following components by weight percentages: 10-50% of CaO, 5-30% of NiO, 5-50% of Al2O3 and 0-30% of SiO2 and 0.5-10% of ZnO can be added. The adoption of the method ensures that the methane process gas that has less than 200ppm hydrogen sulfide can be directly taken as a raw material for reforming the methane steam, which avoids the poisoning of H2S to nickel-based catalyst, reduces the pre-treatment process of the sulfur-containing methane, shortens the process and lowers the cost.

The invention discloses a production method for Pichia pastoris expression recombinant human interleukin 11, which comprises the step of combined purification sequentially through reverse phase column chromatography, hydrophobic column chromatography and gel column chromatography. By adopting the production method, high-purity medicinal recombinant human interleukin 11 proteins can be prepared ina large scale.

The invention discloses a composite silica gel column and a pretreatment method for simultaneously measuring multi-type persistent organic contaminants in sediments and biological samples. The composite silica gel column comprises glass wool, silver silicate silica gel, activated silica gel, alkaline silica gel, activated silica gel, first acid silica gel with the acid content of 40 wt%-50 wt%, second acid silica gel with the acid content of 20 wt%-25 wt%, activated silica gel and anhydrous sodium sulfate sequentially from bottom to top.