36results about How to ""adaptive" patented technology

The invention provides a tool abrasion monitoring method based on wavelet denoising and Hilbert-Huang transformation. According to the method, on the basis of wavelet denoising, the Hilbert-Huang transformation method is used for analyzing signals acquired in the machine tool operating process; according to an amplitude average value, obtained through EMD, of each IMF component and a time-frequency spectrum and a marginal spectrum which are obtained through Hilbert transformation, a normal state of a tool and an abrasion state of the tool are compared, and obvious characteristics are obtained. The self-adaptability of the method is much higher than that of other characteristic extraction methods; according to the method, the defects of high-frequency resolution are overcome, and meanwhile signal characteristics further retain all physical significances contained in original signal characteristics.

The invention discloses a dynamic data scheduling method in a multi-path transmission control protocol (MPTCP), comprising the following steps: the round-trip time RTT and the round-trip time variation RTTVAR of each path are measured; whether each path is available is judged according to the RTT and RTTVAR, the congested paths are abandoned, and the available paths are retained; the data transmission capacity of each path is estimated; and a data transmission quota is allocated for each path based on the estimated data transmission capacity and the current sending window, and a scheduler allocates transmission data according to the transmission quota. According to the scheme, the essence of a TCP Vegas algorithm is absorbed on the basis of Lowest_RTT, and improvement is made on the basis, which is equivalent to adding a congestion avoidance control function to the scheduler. The method has far-reaching significance to avoiding path congestion and enhancing the overall transmission capacity of the paths.

The invention relates to a nanometer composite, and particularly relates to a method for preparing a nanometer composite with a chrysanthemum structure. The method comprises the steps of adding titanium dioxide nanosheets into deionized water, carrying out ultrasonic dispersion, and obtaining a titanium dioxide dispersing liquid; under the condition of magnetic stirring, adding sodium molybdate and thioacetamide into the titanium dioxide dispersing liquid, and obtaining a mixed solution A; after finishing adding, stirring the mixed solution continuously, then transferring the mixed solution into a Teflon inner container, sealing the inner container in a stainless steel hydrothermal reaction kettle for hydrothermal reaction, after finishing the hydrothermal reaction, naturally cooling the reaction kettle to room temperature, centrifuging the obtained product, washing, vacuum drying and obtaining the composite. The method has the advantages that the source of raw materials is wide, and the method is simple and practicable and has low cost. The prepared titanium dioxide or molybdenum disulfide nanometer composite not only has good capacity in absorbing organic contaminants, but also has a good visible light photocatalytic degradation effect to organic contaminants.

The invention relates the anthropoid mechanical arm device. The invention solves the problem that self-adaptive artificial hand only can not realize the linkage. The little finger torsion spring (3-10) is connected with little finger torque pick-up (3-7) installed on fulcrum (6), the big bevel gear (4-3) fixed on fulcrum (6) is contacted with bevel pinion (4-8) fixed on decelerator (4-11), decelerator (4-11) is connected with motor (4-12), ring finger torsion spring (4-6) is connected with ring finger torque pick-up (4-5) and middle finger spring support (5-6) fixed on fulcrum (6), and the middle finger torsion spring (5-9) is connected with middle finger spring support (5-6) and middle finger torque pick-up (5-8) fixed on fulcrum (6). The structure can realize finger linkage.

The invention relates to an image super-resolution reconstruction method based on sparse representation and adaptive filtering. A lot of images are fully clustered by utilizing structure information of image contents firstly, each kind of image set is guaranteed to contain high-consistency image structure information, sparse representation dictionaries of categories are obtained through principal component analyses, performed on the basis, by the categories, and the adaptability is high. Through adoption of a packet minimum angle regression method, an Euclidean projection method on an l<1)-ball, and through a cross iteration mode, high- and low-resolution image block mapping relationship matrixes of each category are solved. Field processing is performed on the low-resolution images directly by utilizing the mapping relationship matrix learned through training finally, and high-definition high-resolution images are rapidly reconstructed.

The invention provides an intelligent material driven or mechanical driven aircraft with turnable wings, and the aircraft provided by the invention has the advantages of high speed, long voyage, low oil consumption, capability of VTOL (vertical take-of and landing), and capability of hovering. The aircraft comprises an aircraft body, sectional bending turnable wings, electroactive polymer skins, sheet macro-fiber piezoelectric composite material bars, deformed honeycomb fillers, a reversal platform and variable pitch propeller engines, wherein the two variable pitch propeller engines are symmetrically mounted on the rear edges of the wings, the electroactive polymer skins are taken as wing skins, the sheet macro-fiber piezoelectric composite material bars are arranged on the neutral layers of the wings, the spaces between the wing skins and the sheet macro-fiber piezoelectric composite material bars are filled with the deformed honeycomb fillers, and the reversal platform is placed ata rotary center inside the aircraft body. According to the invention, an intelligent driver is utilized as the driving source for wing turning, and the aircraft has the advantages of light weight, simple mechanism, high efficiency and convenience for maintenance, and can be switched rapidly in flying modes of level flaying and hovering through wing turning.

The invention relates to an intelligent protection center-based wide-area backup protection system and method. The system comprises a GOOSE network for acquiring switching quantities of a plurality of substations, a sampling value network for acquiring electrical quantities of the substations, relaying protection units connected with the GOOSE network and the sampling value network and a protection intelligent center connected with the relaying protection units through a wide area network; the system further comprises merging units which are connected with mutual inductors of the substations; the merging units merge the electrical quantities measured by the mutual inductors and performs synchronization processing on the electrical quantities, and transmit the processed electrical quantities to the relaying protection units through the sampling value network; the switching quantities of the substations are transmitted to the relaying protection units through the GOOSE network; the relaying protection units calculate the direction of a fault through the obtained electric quantities, and transmit the received switching quantities and the direction of the fault which is obtained through calculation to the protection intelligent center through the wide area network; and the protection intelligent center determines the position of the fault and issues a tripping instruction. The intelligent protection center-based wide-area backup protection system and method of the invention are unrelated to the structures of power grids and the types of faults, and have the advantages of high movement speed, high reliability and strong adaptability.

The invention discloses a novel automatic selection method of an optimal segmentation scale of a high-resolution remote sensing image. A multi-scale segmentation model of a high-resolution image is established by using a multi-scale MRF model, meanwhile image layer segmentation and image plane modeling are carried out at the same time, and context information between layer and layer objects and the spatial dependency of objects in the same layer are described. Spectrums, colors, textures, topological relations and other basic features of the objects are normalized in a Markov random field, a global optimal segmentation scale selection method capable of being automatically executed by a computer is realized by probabilistic information convergence calculation, parameter selection calculation and inference engineering are automatically executed by the computer, and an optimal segmentation scale parameter on theory is obtained. The technique has the advantages of high segmentation quality precision, high self adaptability and high computational efficiency.

The invention discloses a die cutting device of a plastic bag notching machine. The die cutting device comprises a rack, an air cylinder, guide rods, a cutting board, a stamping knife and a stamping knife equalizing adjusting device, wherein the stamping knife equalizing control device is composed of an adjusting board, a stamping knife seat board, an elastic connecting mechanism and adjusting screws; the elastic connecting mechanism is composed of sliding rods, springs and spring bases, and the elastic connecting mechanism is used for causing the adjusting board and the stamping knife seat board to form an elastic connection; and the four adjusting screws are arranged on the adjusting board, the contact angle of the stamping knife and the cutting board is changed through screwing or unreeling the adjusting screws in one direction, and the die cutting barycenter is adjusted. The die cutting device provided by the invention has the advantages that the structure is simple, stable and reliable, and the adjustment is convenient and rapid.

The invention relates to a true random bit stream generator, which comprises a random noise extraction module and a bit stream output module. The random noise extraction module is used for extracting the noise at two ends of a voltage-regulator diode and generating a random digital noise signal to the bit stream output module; and the bit stream output module processes the random digital noise and outputs the true random bit stream. Due to the adoption of a multi-path noise extraction method, the randomness of the output bit streams is further improved, and the generated random data has good statistical property before postprocessing.

The invention relates to an integrated control system and control method for a variable optical attenuator. The system is provided with a configuration management module, an input quantity management module, an attenuation amount voltage lookup table, a first configurable timer, a second configurable timer, a third configurable timer, a table look-up management module, a PID (Proportion Integration Differentiation) input management module, a PID calculating speed management module, a PID calculating module and an output quantity management module. The method comprises the following steps of: starting the configuration management module; starting the input quantity management module; calculating a current actual attenuation amount; entering the table look-up management operation; entering the PID input management operation; calculating a current desired attenuation value and a current actual attenuation value; judging time timed by the second configurable timer; carrying out PID calculation; outputting a calculating result to the output quantity management module; calculating a current table look-up attenuation amount and a table look-up threshold set value; judging whether the current table look-up attenuation amount is less than or equal to the table look-up threshold set value or not; and judging whether time timed by the first configurable timer is up or not. By the integrated control system and control method, the control on different types of variable optical attenuators can be realized.

The invention relates to an adaptive method for classifying communication environments in network planning, and relates to wireless network planning and optimizing technology. The method comprises the following steps of: selecting a sample point to acquire an attribute vector of the sample point; calculating the absolute similarity of a target cell and a cell where the sample point is positioned; calculating the relative similarity according to the percentage pij of the geo-information and building information of the cell oi, in which the sample point is positioned, in a grid; and performing cluster selection of the target cell according to the relative similarity. When the communication environment is changed, the method has good adaptivity and transportability, has good classification effect and precision although the execution speed is in a tolerable range, and can provide important reference for wireless network optimizing and planning.

The invention relates to a dictionary learning algorithm based on sparse model analysis. According to the algorithm, a cost function is established directly through noise signals, the cost function is solved through a gradient descent method, and it is guaranteed that the value obtained through iteration conducted each time is optimal within the local range. A dictionary is adaptively updated through the algorithm, so that a super-complete dictionary capable of better expressing the image structure is formed. The dictionary learning algorithm can be used for image noise removal, image quality can be improved, more accurate target and background information can be provided, and the ideal noise removal effect can be achieved. The algorithm is widely applied to the military field and non-military field of target detection, optical imaging, safety monitoring systems and the like.

The invention belongs to the technical field of energy and power, in particular to an unattended operation type monitoring system and method for organic Rankine cycle isolated network power generation. According to the unattended operation type monitoring system and method for organic Rankine cycle isolated network power generation, automation control of the monitoring system for organic Rankine cycle isolated network power generation and system malfunction automatic processing are achieved. The monitoring system comprises a flowmeter, pressure sensors, temperature sensors, a working medium pump, an evaporator, an electric three-way valve, an expansion machine, a power generator, a condenser, a heat source circuit, a cooling circuit, a touch screen, a PLC unit, a working medium pump frequency converter, a relay, a warner and a multifunctional electrodynamic meter. Through real-time state monitoring of a unit, under instable heat-source and cold-source fluctuation, stabile, reliable and efficient operation of a waste heat power generation system is achieved. According to the unattended operation type monitoring system and method, while good self-adaption is achieved, automatic adjusting can be conducted when the system malfunctions, and unattended operation of an isolated network power generation system is achieved.

The invention discloses a digital image stitching tampering blind detection method, which respectively utilizes an LMD (Local Mean Decomposition) method and a CCDCT method to extract image features, wherein the LMD method decomposes original signals into the sum of a plurality of PF components, the PF components are obtained by multiplying envelope signals and pure frequency modulation signals, the envelope signals represent the instantaneous amplitude of the PF components, and instantaneous frequency is obtained by the pure frequency modulation signals. Accordingly, the instantaneous amplitude and instantaneous frequency of all PF components form the complete time-frequency distribution of the original signals; based on this, feature extraction can reflect influence caused by stitching operation to a certain degree. Stitching operation may influence image features such as smoothness, conformity, continuousness, regularity, etc. so as to change inner links among pixels; a series of two-dimensional arrays can be obtained by utilizing the CCDCT method to reflect feature changes, so that the CCDCT method can extract features and effectively stitching images; features extracted by two methods are used as mixing features to verify image authenticity combined with an Adaboost classifier, and better effects can be obtained.

The invention relates to an azimuth ambiguity suppression method for along track interferometry SAR ocean current inversion, and belongs to the technical field of ocean remote sensing. The method comprises the following steps of: converting an along track InSAR two-channel SAR complex image corresponding to a certain region of a sea surface which has ocean currents to be measured to range frequency-Doppler frequency domain after registration; calculating a covariance matrix of an along track InSAR by using independent identical distribution samples of a range frequency dimension; calculating acurve of along track interferometry characteristic spectrum entropy which changes with Doppler frequency according to the calculated covariance matrix; according to the curve, finding maximum pointsin a left half part and a right half part of the curve separately; then removing all Doppler units at the left side of the left side maximum points of the curve and removing all Doppler units at theright side of the right side maximum points of the curve; determining a threshold value of the characteristic spectrum entropy by using an adaptive iterative method; removing Doppler units with the characteristic spectrum entropy greater than the threshold value; and calculating range velocity of the ocean currents by using an interference method based on obtained InSAR data.

The invention discloses a multi-stream feature distance fusion system, which comprises a multi-stream feature extraction network, a contribution coefficient adaptive generation module and a distance fusion module which are connected step by step, wherein the multi-stream feature extraction network comprises a feature map extraction network used for extracting feature maps of input images and a feature extraction network used for extracting corresponding features from the feature maps respectively; the contribution coefficient adaptive generation module comprises an activation ratio calculationmodule and a contribution degree mapping module for calculating contribution coefficients of the regional characteristics according to the activation ratio of the regional characteristic patterns; and the distance fusion module is used for calculating feature distances between the corresponding feature maps and fusing the feature distances into multi-flow feature distances by utilizing contribution coefficients of the features. The invention also discloses a multi-stream feature distance fusion method. According to the method, the reference index which is more in line with the real conditioncan be provided for image similarity judgment through the multi-stream feature distance, and a breakthrough progress is brought to improving the accuracy of image similarity judgment.

The invention discloses a beam-end connecting device of a novel structure and provides a connecting device capable of resetting easily, replaceable, resistant to bending and capable of dissipating energy. The connecting device is applied to the beam-end portion of a framework beam or a coupling beam in a building structure or a bridge structure, the connecting device is activated when the beam-end generates bending moment to a certain degree, and a damper of the connecting device can remarkably improve the structure damping level. Under the condition that the structure bears general power loads, the beam-end connecting device of the novel structure provides damping, thereby assisting in controlling storey drift, speed and acceleration. Under the condition that the structure bears high-intensity power loads, the connecting device has self-adaptability to a certain degree, after the storey drift angle is larger than a preset critical value, the connecting device can centralizeearthquake damage to a replaceable intensity weakening section of the connecting device, ductility and energy dissipation are provided through buckling deformation of flanges, and the overall stability of the structure is ensured. Besides, the connecting device is replaceable and provided with a rotation reset button, thereby restoring post-earthquakeresidual drift easily, being convenient to replace and effectively reducing all kinds of losses caused after an earthquake.

The invention belongs to the technical field of image conversion and compression, and discloses an image conversion method. The image conversion method comprises the following steps: 101: analyzing an original color image to obtain the color information of all information contents in the image; 102: carrying out weighted processing on the color information, and converting the weighted color image into a grayscale image; 103: analyzing the grayscale image to obtain the resolution of the grayscale image, obtaining an adaptive local range under the resolution, and determining the size of an area impact factor; and 104: traversing the whole image according to the size of the area impact factor, and converting the image into a black-and-white image through an adaptive image binaryzation method with a limited range. By use of the method, the image is clearer after being converted, and is easy in optical recognition.

The invention discloses a rehabilitation manipulator based on high-order adaptive learning control. The rehabilitation manipulator comprises a plurality of rehabilitation mechanical finger joints anda processor; and the processor comprises a first acquisition module, a second acquisition module, a third acquisition module and a control module. According to the technical scheme, the rehabilitationmanipulator can be used for completing tracking tasks on respective expected tracks after a certain number of times of learning, the control problem is solved, and the control precision is high.

The invention discloses a method and a device for correcting capital and lowercase forms of characters in western language words and belongs to the field of word processing. The method comprises the following steps of: acquiring identification results of the characters of a target text and positions of the characters in an image; judging base line attributes of the characters according to the identification results of the characters and the positions of the characters in the image, and determining the base line information of the words after counting; and determining the base line attributes of the characters according to the base line information of the words, correcting the base line attributes of the characters, and outputting the identification results. By the method, the words which are not accordant with an ordinary writing specification or cannot simply count to obtain base lines are effectively processed, and the capital and lowercase forms of the characters can be effectively processed and determined after counting the base line attributes of the characters obtained under the condition that a word row warps. The invention has excellent adaptivity and greatly improves the identification rate of the words.

The invention provides an electromagnetic target intelligent perception identification method based on multi-feature heterogeneous fusion, and the method comprises the steps: firstly obtaining electromagnetic target signal source data information through a sensor, then carrying out the preprocessing of data through heterogeneous data extraction and multi-feature fusion, constructing a target template through a data sample, and building a target knowledge base; then, electromagnetic target sensing recognition based on a machine learning strategy is used for completing radiation source recognition, a recognition result obtained through a transfer learning strategy is stored in a target knowledge base, and a result obtained through an online learning strategy is used for updating the knowledge base. According to the method disclosed by the invention, the identification capability of the electromagnetic target is improved by more accurately describing the characteristics of the electromagnetic target, and online learning, transfer learning and the convolutional neural network are combined, so that the identification speed of the electromagnetic target and the robustness of a system to scene and sensing equipment transformation are improved.

The invention relates to the technical field of electronic test device alarming, and discloses a real-time alarming method and device for the working condition of an electronic test system. The real-time alarming device adopted by the real-time alarming method comprises a test data interface device, a stack-type memory for storing test data and an alarming information processing system for processing the stored test data; the test data interface device is connected with the stack-type memory through a data transmission cable; the alarming information processing system is used for processing the data stored by the stack-type memory; and the processed test data result is compared with a threshold value to obtain judgment information, and the judgment information is transmitted into an alarming device through the data transmission cable. Subjective experience factors of people can be overcome, and the alarming confidence level can be provided; the working sate of an electronic test devicecan further be reflected to the largest limit in real time, and supersession of the old by the new is continuously conducted; and the difficulty that with increasing of the test data, the modeling operation number is continuously increased can further be avoided.

The invention discloses a CT image denoising method based on wavelet transformation, belongs to the technical field of medical image processing, and is particularly suitable for CT image denoising of new crown pneumonia. The CT image is susceptible to the interference of Gaussian noise in the transmission and acquisition process, and the wavelet transform can effectively remove the interference of the Gaussian noise. In order to solve the problems that early-stage lesions of a new crown CT image are not obvious in change, the number of the lesions is small, the range of the lesions is small, the density is low, and missed diagnosis of early-stage new crown patients is easily caused, the contrast ratio of the new crown lesions is improved, namely, an arc tangent improved self-adaptive wavelet threshold function of index adjustment and an improved threshold based on contraction factors are provided; the arc tangent function changes quickly near the zero point and changes slowly away from the zero point, the exponential function is adjusted to adapt to different layer threshold functions, more high-frequency detail information in the lung CT image is obtained, the detail edge is reserved, and fuzziness is reduced. The selection of wavelet threshold function parameters is a key factor for determining distortion and errors after image denoising, and the optimal adjustment parameters are searched through the improved particle swarm optimization of sine and cosine fusion normal distribution, so that the threshold optimization effect is greatly improved.

The invention relates to a compression roller self-adaptive compression structure, an ultrasonic machining device and a mask machine, and belongs to the technical field of effective conveying or machining of to-be-machined workpieces by applying pressure through compression rollers in cooperation with pressing. The compression roller self-adaptive compression structure, the ultrasonic machining device and the mask machine are particularly suitable for conveying and machining of the to-be-machined workpieces which are of a compressed multi-layer structure, for example, the ultrasonic machiningdevice used in the mask machine achieves effective conveying, welding and other machining of to-be-machined masks by using the compression rollers in cooperation of pressing. The compression roller self-adaptive compression structure is provided with a supporting structure, the compression rollers and a self-adaptive pressing device, and the self-adaptive pressing device applies working pressure to the compression rollers and achieves self-adaptive floating adjustment along with the change of the contact state of the compression rollers. The compression roller self-adaptive compression structure has the following beneficial effects that the compression roller self-adaptive compression structure can achieve self-adaptive adjustment, meets intelligent control, is convenient to operate, is applied to the ultrasonic machining device and the mask machine, improves the production efficiency and the production quality of the ultrasonic machining device and the mask machine, is low in workingnoise and small in occupied area, and conforms to industrial utilization.

The invention discloses an aircraft taxiing scheduling method based on taxiing conflict risk prediction, and the method comprises the steps: building a kinematic model according to the maximum taxiing speed and the maximum running speed of aircrafts of different types; an aircraft vectorization course model and an aircraft taxiing conflict risk assessment module are established, so that the time for entering conflict risk prediction is more accurately judged; the taxiing distance of the aircraft driven by a driver on a certain taxiway basically conforms to the truncated normal distribution in the taxiing process of the aircraft, and a taxiing conflict risk prediction method is established; by accurately judging the conflict risk prediction time and accurately predicting the position probability of the aircraft, the conflict risk of the aircraft is evaluated and predicted, and the prediction and evaluation precision is improved; and three control methods for reducing the conflict risk are established according to the sliding conflict risk prediction value.

The invention provides a heat preservation cover for wrapping tropical plants for overwintering. The heat preservation cover comprises a fixed support, a heat preservation cover annular storage cavity, inflation keels and a heat preservation layer, wherein the fixed support is an annular support, and through hinges arranged at the end of the fixed support, the whole annular support can be folded in half and expanded; the heat preservation cover annular storage cavity is a stainless steel hollow annular cavity, an upper cover is further arranged on the heat preservation cover annular storage cavity, and the inflation keels and the heat preservation layer are placed in the heat preservation cover annular storage cavity; each inflation keel is an inflatable cylinder made from an oxford fabricmaterial, the heat preservation layer is made of a dual-layer oxford fabric material and used for wrapping the plants, the inflation keels are arranged on the heat preservation layer at intervals, and a valve core is further installed in the heat preservation cover annular storage cavity, connected with all the inflation keels and used for inflating the inflation keels after being connected withan inflator. By means of the heat preservation cover, the problem in the prior art is solved that overwintering wrapping technologies for tropical plants in the prior art have large difficulty and high cost.

The invention discloses a continuous firing mechanism of a paintball gun. The ideal full-automatic continuous firing of the paintball gun is achieved. The continuous firing mechanism of the paintball gun comprises a sear, a driving hammer, the driving hammer drives counter-recoil through a driving hammer spring, the continuous firing mechanism of the paintball gun further comprises a delay body, the delay body is arranged on a counter-recoil spring guide rod in a sliding fit mode and drives counter-recoil through a delay body counter-recoil spring arranged on the counter-recoil spring guide rod in a sleeving mode, a running-firing sear is hinged to the sear, a continuous firing sear drives counter-recoil through a continuous firing sear spring, a strip-shaped groove is formed in the lower end of the driving hammer, the free end of the continuous firing sear is provided with a blocking part protruding upwards and a pressing part, a receding space is formed between the blocking part and the pressing part and used for receding the lower portion of the rear end of the driving hammer, the front end of the blocking part is provided with an inclined face used for being matched with the rear end of the driving hammer to press the free end of the continuous firing sear downwards, the rear end of the blocking part is provided with a limiting face used for being matched with the rear wall of the strip-shaped groove to limit the driving hammer, and the rear end of the pressing part is provided with the inclined surface which is used for being matched with the front end of the delay body to press the free end of the continuous firing sear downwards, so that the limiting of the driving hammer is relieved.

The invention discloses a forest growth modeling and predicting method and device based on machine learning, and relates to the field of forest industry. The method comprises the following steps: collecting destructive sampling image data of a tree through at least one image sensing device; labeling annual rings in the image data to obtain labeled data; obtaining basic parameters of a tree corresponding to the image data, wherein the basic parameters at least comprise tree biological quality data; inputting the image data, the annotation data and the basic parameters as training data into a machine learning model, and performing training to generate a biological quality model, wherin the biological quality model is used for automatically identifying the subsequently collected destructive sampling image data and outputting current and predicted tree biological quality data. By adopting the technical scheme of the invention, the multi-dimensional training data can be input into the machine learning model to generate the biological quality model, so that the biological quality model is higher in adaptability and higher in prediction accuracy.

The invention relates to an expansion type interbody fusion cage. The expansion type interbody fusion cage comprises a rectangular frame, wherein an upper cover plate with an upper window is installed at the top of the frame, and a lower cover plate with a lower window is installed at the bottom of the frame; an n-shaped upper expansion plate and an n-shaped lower expansion plate are arranged in the frame, the middle of the upper expansion plate falls into the upper window, the width of the upper window is greater than that of the middle of the upper expansion plate and less than the distance between the outer side edges of a left upper folded plate and a right upper folded plate of the upper window, the middle of the lower expansion plate falls into the lower window, the width of the lower window is greater than that of the middle of the lower expansion plate and less than the distance between the outer side edges of the left lower folded plate and the right lower folded plate; and a back-up roll which can be rotationally adjusted from the outside of the frame and has an oval section is mounted in the middle of an inner cavity of the frame, and the surface of the back-up roll is in contact with the inner walls of the upper expansion plate and the lower expansion plate. The expansion type interbody fusion cage can better adapt to the angles of cartilage end plates on the two sides, the contact area between the device and the cartilage end plates is larger after the device is fully expanded, and stability is improved.