73results about How to "Resolve mismatch" patented technology

The invention discloses a dynamic environment dense point cloud SLAM (Simultaneous Localization and Mapping) method and system based on YOLOv11 and ORB-SLAM 3. According to the method, real-time target detection and image segmentation technologies of a YOLOv11 model are integrated into an ORB-SLAM 3 framework, so that high-precision and robust visual SLAM in a dynamic environment is realized. A traditional convolution layer is replaced by using a balanced convolution method GSConv layer in a YOLOv11 model, and a traditional C2f layer is replaced by using a new feature fusion module VoVGSCS layer, so that a Neck structure of the YOLOv11 is improved, and a lightweight network model is realized. Experimental data prove that the pose estimation precision of the method in a dynamic environment is remarkably superior to that of an existing visual SLAM algorithm.

The invention relates to the technical field of data processing, and discloses a vehicle-mounted head-up display visual focus prediction method and device, equipment and a medium, and the method comprises the steps: collecting an activity video of eyeballs of a user; determining a historical movement track of the sight focus based on the activity video, and calculating the delay time of the vehicle-mounted head-up display system according to the historical movement track; predicting a screen projection position corresponding to the sight line focus of the next frame according to the delay time to obtain predicted coordinate information; and controlling the vehicle-mounted head-up display system to execute projection operation according to the predicted coordinate information. The problem that the screen projection position is not matched with the actual visual focus position of the driver due to delay in the AR-HUD system in the existing eyeball tracking technology is solved.

The invention belongs to the technical field of intelligent digital signal processing, and discloses a speaker privacy protection method and system capable of retaining speech emotion, and the method comprises the steps: obtaining the original audio of a speaker, and inputting the original audio into an end-to-end emotion speech fusion model; reasoning the original audio by using the fusion model, changing voiceprint information in the original audio, and obtaining a pseudo speaker audio with an emotional state reserved; the fusion model comprises a prior encoder and a waveform decoder, the prior encoder receives an original audio and a speaker identity identification number ID, and calculates and uses a fundamental frequency code, a content code, an emotion code and a speaker ID code of the original audio to generate a prior distribution mean value and a variance; and the waveform decoder restores the prior distribution mean value and variance into pseudo speaker audio. And a model fusion means is used, so that the unseen pseudo speaker audio is generated, the problem of "out-of-distribution" in the reasoning process is avoided, and the problem of mismatching between emotion retention and pseudo speaker information is solved.

The invention relates to the technical field of laser communication, in particular to a free space optical communication system, device and method. According to the free space optical communication system provided by the invention, the space light beam is firstly coupled with the multimode optical fiber, and the characteristic that the diameter of a fiber core of the multimode optical fiber is larger is utilized, so that the coupling efficiency is improved, and the energy loss in the coupling process can be effectively reduced. Then, a mode demultiplexer is used, and a mode conversion technology is adopted to separate and convert multiple modes of light in the multimode optical fiber, so that the obtained LP01 mode is separated and converted into a fundamental mode of the single-mode optical fiber to serve as an output optical signal, and the problem that the modes of the multimode optical fiber and the single-mode optical fiber are not matched is solved through the mode conversion technology; and finally, the converted fundamental mode signal in the single-mode fiber is used for communication, so that a high-speed communication module can be better matched.

The invention relates to the technical field of computer vision, in particular to a NeRF map construction method and device based on visual SLAM, and the method comprises the steps: obtaining a static target feature point of a current frame image of a camera in a dynamic scene; tracking the static target feature points, solving the current frame camera pose of the camera, and selecting a key frame according to the number of projection inner points and the tracking time in the tracking process; pixel point sampling is conducted in the key frame according to ORB feature guidance, a key frame sampling result is input into a pre-established NeRF map model for training, NeRF map model parameters are obtained, and a NeRF map construction result is obtained. According to the method, the dense NeRF map can be generated in real time for the dynamic scene, the problem of mismatching in the dynamic scene is solved, simplicity is kept while high efficiency is achieved, the influence caused by a dynamic target is eliminated, high-quality background reconstruction is completed, and the positioning and mapping application requirements in different scenes are effectively met.

The application discloses a kind of full-process supervision systems based on visual offline identification, and the application relates to component production technical field, it is solved that existing technology mostly stays in single size comparison level, lack of the problem of full-process integrated supervision mechanism from model generation, standard matching, automatic splicing to assembly verification, the unified check feature of the present application is constructed by built-in midpoint, plane midpoint and feature vector, corresponding three-dimensional standard model can be matched from model database quickly and accurately, the identification logic is stable, and the anti-interference is strong, effectively avoid the model mismatching problem caused by similar appearance;Through the vector comparison of reference feature and component feature, length difference and included angle calculation, the assembly accuracy can be quantitatively judged, and the assembly standard, correctable and substandard state can be distinguished, the visualization and digitization of assembly quality are realized, the on-site quick error correction and traceability are facilitated, and the assembly reliability and overall production quality of flow line are effectively improved.

The utility model discloses an intelligent scale for fresh Chinese wolfberry fruits. The intelligent scale comprises a weighing assembly, an identification assembly and a display assembly, the weighing assembly comprises a scale body and a gravity sensor, and the gravity sensor is embedded into the upper surface of the scale body; the identification assembly comprises a first camera and a gravity trigger switch, the gravity trigger switch is cooperatively connected to the lower part of the gravity sensor, and the first camera is connected to one side of the scale body through a supporting rod and is electrically connected with the gravity trigger switch; the display assembly is electrically connected with the gravity sensor and the first camera. The first camera is used for photographing, the gravity sensor is used for weighing, and the images are displayed on the display assembly, so that a picker can simply and directly see himself or herself and the picking weight of the picker, the problem that the picking weight is not matched with the picker is solved, and the problems of manual recording omission, wrong writing and the like are solved; the photographing process of an operator is reduced, the working efficiency is improved, and the operation is more convenient.

This utility model relates to the field of medical device technology and discloses a nasogastric tube feeding pump connector. The connector includes a connector body, a feeding tube connector mounted at the top of the connector body, and a screw-type nasogastric tube connector at the bottom of the connector body. A fixing plate is provided on the outer wall of the connector body. This utility model, by setting a straight-mouth nasogastric tube connector, a screw-type nasogastric tube connector, and a connecting block, allows the straight-mouth nasogastric tube to be connected to the straight-mouth nasogastric tube by threading the connecting block, thus connecting the straight-mouth nasogastric tube to the feeding tube. By unscrewing the straight-mouth nasogastric tube connector from the connecting block, the screw-type nasogastric tube can be directly connected to the screw-type nasogastric tube connector, thus connecting the screw-type nasogastric tube to the feeding tube. This solves the common clinical problem of mismatch between nasogastric tubes and feeding tube connectors, and simultaneously meets the needs of connecting screw-type and straight-mouth nasogastric tubes to nasogastric feeding tubes.

The positioning method, device, robot, medium and product of the robot provided in the application, the method comprising: obtaining a current map generated by the robot at a current position, matching the current map with a preset global map to obtain a matching result; the matching result comprising at least one candidate pose solution and a matching degree corresponding to the at least one candidate pose solution; in a case where it is determined that the matching result satisfies a first preset condition, controlling the robot to perform a subsequent task, and in the process of performing the subsequent task, verifying the at least one candidate pose solution based on an updated current map obtained to obtain a target pose solution. The application quickly generates candidate results through local data and dynamically verifies during cleaning, effectively shortens the repositioning time, and improves the user experience.

The invention discloses a lane signal map construction method and system based on crowdsourcing data, and belongs to the technical field of vehicle localization, and the method comprises the steps: employing a Bayesian state estimation algorithm to estimate the obtained crowdsourcing data, and obtaining a vehicle motion track, inertial navigation error model parameters and magnetometer error model parameters; carrying out feature representation on the magnetic field sequence based on the vehicle motion trails by adopting a deep representation learning method to obtain magnetic field sequence features of key frames, and carrying out key frame association on the magnetic field sequence features by utilizing a loopback detection method to obtain loopback constraints among different trails; and constructing a three-dimensional trajectory inequality constraint based on key frame association, and constructing a global trajectory optimization constraint and obtaining a lane signal map in combination with inertial navigation error model parameters, magnetometer error model parameters and loopback constraints among different trajectories. According to the invention, the method achieves the construction of the lane-level road feature map based on crowdsourcing data, and remarkably reduces the cost of a lane-level positioning scheme based on road feature matching.

The present invention discloses a connecting wire clamp hardware and an operation method thereof, wherein the connecting wire clamp hardware comprises: a hardware body; a first wire clamping clip, which is arranged at one end of the hardware body; a second wire clamping clip, which is movably connected to the hardware body and arranged opposite to the first wire clamping clip; a wire clamping bolt, which passes through the first wire clamping clip and the second wire clamping clip, and one end of the wire clamping bolt with a nut is connected to a mechanical arm; the mechanical arm drives the wire clamping bolt to rotate, so that the second wire clamping clip moves back and forth along the side wall of the hardware body to clamp and tighten the main conductor between the first wire clamping clip and the first wire clamping clip. The present invention can be used in conjunction with the mechanical arm of a live robot to complete the live overlap of the main conductor and the drainage line, greatly improving the safety and work efficiency of non-power-off operations.

This invention discloses a method and system for UI element localization in cross-platform, full-scenario RPA based on functional semantics, belonging to the field of robotic process automation technology. It includes a pre-construction step, an element recording step, and an element playback step. The pre-construction step involves pre-constructing a UI element functional semantic ontology library specific to the RPA business scenario. This ontology library includes a standardized element functional semantic tag system independent of the client side, business context rules corresponding to each semantic tag, and interactive behavior features. By pre-constructing a client-side independent functional semantic ontology library specific to the RPA business scenario, and using business functional semantics as the core, this invention constructs a cross-platform generalized feature vector for target elements, freeing them from the binding limitations of recording client features. After the same business process is recorded on the Web client, it can be directly run on PC desktop clients, mobile mini-programs, domestic terminals, and closed systems without interfaces without modification, effectively solving the problem of high repetitive development costs in cross-platform scenarios.

The utility model discloses a power device double-string crimping structure for a converter valve, which relates to the field of flexible direct-current power transmission and comprises an IGBT (insulated gate bipolar transistor) voltage string and a diode voltage string, the IGBT voltage string is composed of a plurality of IGBTs and a first radiator which are stacked in a staggered manner, and the diode voltage string is composed of a plurality of diodes and a second radiator which are stacked in a staggered manner. The IGBT voltage string and the diode voltage string are parallel to each other and are connected through a supporting structural member to form a double-string crimping structure, the second radiator is provided with a protruding structure, the protruding structure is located on the outer side of the bottom of the double-string crimping structure, the first copper bar and the voltage-sharing resistor are installed on the protruding structure, and the second copper bar and the voltage-sharing resistor are installed on the protruding structure. The second copper bar is installed between the first radiator and the second radiator and corresponds to the installation position of the first copper bar, the third copper bar and the laminated busbar are sequentially installed on the side face of the double-string crimping structure in an overlapped mode, the third copper bar corresponds to the installation position of the first copper bar, the pipeline assembly is connected with the adjacent first radiator, and the pipeline assembly is connected with the second radiator. And the adjacent second radiators are connected.

This invention relates to a compliant grasping method and system for robotic arms that integrates RGB-D vision and dynamic inertia adaptive MPC, belonging to the field of robotic arm control technology. It calculates the Cartesian equivalent inertia of the robotic arm's end effector in real time using the joint space inertia matrix and the end effector position Jacobian matrix, constructing a three-layer control architecture of "perception-optimization-execution." It integrates RGB-D visual positioning and multi-stage fifth-order polynomial interpolation trajectory generation technology to achieve a synergistic guarantee of high precision and compliant force control in grasping tasks. Dynamic iterative optimization is performed based on real-time status feedback, ensuring millimeter-level trajectory tracking accuracy of the end effector while strictly limiting the end force within a safe threshold range. This invention achieves deep integration of dynamic inertia estimation and multi-level control, ensuring that the robotic arm can still achieve high-precision and highly compliant grasping operations under complex conditions such as unknown loads and dynamic joint configurations.

The invention provides a pixel gradient and NCC-based wafer image stitching method and device. The method comprises the steps of determining edge pixel points of which the pixel gradient magnitude of an overlapping region of two adjacent line scanning images is greater than a first threshold value; calculating the ratio of the horizontal projection and the vertical projection of the edge pixel points to the height and the width of the unit image; when the ratio is greater than a second threshold value, determining that a graphic feature exists; determining a region of interest according to the graphic features; selecting a region corresponding to the region of interest in the first line scanning image as a template image; determining a credible direction of the region of interest relative to the template image according to the gradient direction and the amplitude of the edge pixel point; performing NCC calculation on the region of interest along the credible direction to determine an integer pixel matching position; sub-pixel offset is calculated through quadratic surface or quadratic curve fitting; and splicing the first line scanning image and the second line scanning image according to the sub-pixel offset. According to the embodiment of the invention, the image splicing precision can be improved while the production efficiency is improved.

This invention relates to the field of three-dimensional image data processing technology, and particularly to a method and system for wire connection node matching and conductor reconstruction based on multi-source point cloud constraints. The method and system, by comprehensively utilizing conductor point cloud data, tower point cloud data, and tower 3D model data, achieve automatic identification and accurate reconstruction of conductor connection relationships between adjacent towers. It fully utilizes the real spatial morphology reflected by the conductor point cloud to establish a globally optimal matching relationship among candidate wire connection node sets of adjacent towers, and integrates structural priors and engineering constraints to accurately determine the actual conductor connection position and achieve continuous spatial reconstruction of the conductor. This solves the problems of low accuracy, poor robustness, and insufficient engineering applicability in existing technologies for conductor connection node matching.

This application discloses an adaptive configuration method, apparatus, device, and storage medium for JP cabinets, belonging to the field of JP cabinet load configuration technology. This application obtains a main node graph characterizing the real-time operating status of a distribution transformer area and a secondary node graph characterizing the future operating trend and adjustability of the distribution transformer area; it performs cross-graph feature fusion on the main node graph and the secondary node graph, and outputs adaptive configuration instructions for the nodes where each JP cabinet in the distribution transformer area is located based on the cross-graph feature fusion result; it sends the adaptive configuration instructions to the corresponding JP cabinets for execution and collects the operating status data after execution; based on the operating status data, it synchronously updates the main node graph and the secondary node graph, and continuously optimizes the configuration parameters of each JP cabinet based on the updated main node graph and secondary node graph, thereby achieving the effect of accurately allocating the load of the JP cabinets.

The present application belongs to copper pyrometallurgy technical field, specifically discloses a kind of dynamic optimization method of double-side blowing copper smelting slag type based on molten pool working condition real-time monitoring and model driving.The method includes: real-time acquisition molten pool temperature, flue gas oxygen content, slag Fe3O4 Content and viscosity and other multiple key working condition parameters;Parameter is input into the preset control model, the expert rule based on metallurgical thermodynamics and production data is embedded in the model, and the dynamic adjustment amount of SiO2, CaO, MgO three kinds of slag forming agent is calculated accordingly;According to the adjustment amount, the proportion of slag forming agent is adjusted in real time;Periodic acquisition of slag sample analysis, compare measured value with model expected value, and feedback correction model parameters, realize closed-loop optimization.The present application can dynamically adjust slag type according to molten pool real-time working condition, reduce the fluctuation of slag fluidity by more than 40%, stabilize the copper content of slag below 1.5%, and improve the direct recovery rate of copper to more than 97.5%, significantly improve the process stability and metal recovery rate.

A binocular vision stereo matching method and storage medium based on regional feature extraction, wherein the binocular vision stereo matching method includes: obtaining a first image and a second image under two viewpoints; setting a first pixel point in the first image, expanding other pixels distributed around the first pixel point with similar textures to form a first texture area, and obtaining a third image using the first texture area and the unexpanded area outside; setting a second pixel point in the second image, forming a second texture area by expanding similar textures and obtaining a fourth image; cost aggregation of any pixel point in the third image or the fourth image according to a preset cost function and a plurality of preset disparity values, so as to obtain the best disparity value of the pixel point from each disparity value. This method can overcome the disadvantage that the low texture content matching process between the two images is prone to errors, and greatly improve the performance of the binocular vision stereo matching algorithm.

The invention relates to the technical field of rolling bearing fault diagnosis, and discloses a full-life rolling bearing fault diagnosis method based on mechanism data migration. The method comprises the steps that a rolling body radial displacement excitation function, the time-varying contact rigidity of a rolling body and a bearing raceway and the impact force at the contact defect position of the rolling body in a mass-spring-damping model are corrected based on the Hertz contact theory, and a nonlinear kinetic equation of the rolling bearing is obtained; solving a nonlinear kinetic equation of the rolling bearing through a fourth-order Runge-Kutta method, generating source domain data consistent with test parameters of the full-life rolling bearing, and constructing a source domain data set with a complete fault tag; according to the method, the dynamic mechanism model consistent with the full-life test bearing parameters is constructed, the feature distribution difference between the generated source domain data and the target domain data is small, a foundation is laid for efficient knowledge migration, and the problem that traditional source domain data is not matched with actual data is solved.

Disclosed is an antigen targeting, anti-CD16A, and immune effector cell activating trifunctional fusion protein, comprising: a CD16A binding region that specifically binds to CD16A, a tumor-associated antigen (TAA) binding region that specifically binds to a TAA, an IL-15 / IL-15Rα complex formed by combining IL-15 and IL-15Ra, and an Fc domain. A designed tumor-targeting innate immune cell activation molecule, i.e., an anti-TAA, anti-CD 16A, and cytokine IL 15 trifunctional fusion protein targets NK cells to a tumor by simultaneously binding to a TAA on tumor cells and CD16A on NK cells, so that the NK cells release perforin and granzymes to induce apoptosis and cause tumor cell death. The cytokine IL15 expands and maintains the function of the NK cells, stimulates immune cell proliferation, and changes the immune microenviroment of the tumor.

The utility model provides a quick plug-in direct current connection structure and a flow battery stack system applicable to the same. Each electric pile comprises an end plate and a collector plate, the rapid plugging direct current connection structure comprises a plurality of sockets, at least one part of each socket is located on the end plate, each socket comprises a collector conducting strip and an insulated connection part, and the collector conducting strips are connected with the collector plates; and a plurality of plug connecting pieces, each plug connecting piece comprises two plug parts and a direct current cable connected between the two plug parts, the plug parts are suitable for being inserted into the connecting parts of the sockets and electrically connected with the current collecting conductive sheets, and the direct current cables comprise flexible and bendable insulated cables. According to the connecting structure provided by the invention, the problem that workpieces are not matched when the galvanic piles in the galvanic pile system of the flow battery are connected can be solved, and the safety is improved while the galvanic piles are conveniently connected.

The invention relates to the technical field of wireless transmission, in particular to an antenna switching circuit and a Wifi radio frequency parameter algorithm design method. Comprising a radio frequency switch, and the radio frequency switch is provided with a radio frequency input port used for receiving Wifi 2.4 G, 5G or 6G signals and a control port used for being electrically connected with a GPIO port of a CPU to receive control signals sent by the CPU so as to switch a radio frequency path. By designing the antenna switching circuit comprising the radio frequency switch, the flexible switching of the Wifi signal among different radio frequency paths is realized, and the design is copied by other links of the MIMO link, so that the consistency of multi-link switching is ensured. In combination with a Wifi radio frequency parameter algorithm, firstly, original calibration parameters are generated for a path with the maximum insertion loss as a basis, then multi-state correction parameters are generated for other paths according to characteristics such as link insertion loss and reflection indexes, and the corresponding correction parameters are called when the radio frequency paths are switched, so that optimal matching of radio frequency parameters of different paths is realized.

The invention relates to the technical field of image processing, in particular to an equipment cluster target information positioning method based on a neural network model, and the method comprises the following steps: 1, obtaining an operation instruction of target equipment and video information corresponding to the operation instruction, and generating an operation sample; collecting a plurality of operation samples for each type of operation instruction, and generating an operation sample data set corresponding to the operation instruction; 2, acquiring image information of an operation panel of the target equipment, and generating three-dimensional modeling information of the operation panel based on a point cloud algorithm; 3, inputting the operation sample data set into the feature extraction network model; and 4, correspondingly mapping the targets in the target instruction set into the three-dimensional modeling information, and generating target mapping information of the operation instruction in the real world. According to the scheme, the intelligent identification and positioning of the equipment operation state and the target information can be realized without carrying out any physical transformation on the equipment or obtaining the authorization of an original factory interface.

The application provides a repeated value sorting method for inequality connection in a database and a computer readable storage medium. The repeated value sorting method for inequality connection in the database comprises the following steps: receiving a query request containing multiple inequality connection conditions in a database, determining connection columns corresponding to each connection condition and a comparison operation type; setting a sorting rule according to the comparison operation type, sorting and merging left table and right table connection columns participating in connection to construct a first sorting array with a repeated value flag bit and a table identification bit; constructing a second sorting array and a permutation array based on the first sorting array, initializing a Bloom filter corresponding to a data block of a bit array; double-pointer traversing the second sorting array to match tuples satisfying the connection condition, setting corresponding positions of the bit array and the Bloom filter as valid identification values; and coarsely filtering the bit array through the Bloom filter and scanning effective data blocks, combining the table identification bit to map the matched tuples back to original tables to obtain an inequality connection result of the database.

The invention particularly relates to a semantic knowledge migration-based high dynamic range imaging method, which comprises the following steps of: acquiring a low dynamic range image, and generating an initial high dynamic range image by utilizing an initial high dynamic range reconstruction model; converting the initial high dynamic range image into a reconstructed image of a standard dynamic range domain through domain migration operation; processing the reconstructed image of the standard dynamic range domain by adopting a semantic segmentation network SAM to obtain semantic priori; processing the reconstructed image of the standard dynamic range domain by combining semantic prior with a semantic prior guidance reconstruction model to obtain a refined reconstructed image; based on a knowledge distillation method, cross-domain semantic priori migration is realized at three levels of content, color and features, and an initial high dynamic range reconstruction model is optimized through a combined loss function; in the reasoning stage, only the optimized initial high dynamic range reconstruction model is adopted to output a final high dynamic range image. According to the method, high-dynamic-range image reconstruction is carried out by effectively utilizing the advanced capability of semantic priori.

This invention relates to the field of marine power system control technology, and particularly to a hybrid power management method for ships based on the synergy of shaft-driven power generation and lithium batteries. The method includes: real-time acquisition of the ship's power grid load demand, the output power of the shaft-driven power generation system, the state of charge of the lithium battery system, and the rotational speed of the ship's main engine; adaptively determining the current system operating mode; calculating the real-time power difference between the load demand and the output power of the shaft-driven power generation system; generating a target charge / discharge power for the lithium battery system; obtaining a safe command charge / discharge power; controlling the lithium battery system to execute the command charge / discharge power, and controlling the shaft-driven power generation system to operate according to the target of the current mode. Through precise power scheduling and adaptive switching of system operating modes, the method ensures that the ship's power system always operates in an energy-optimal state under different operating conditions.

The invention discloses a security check method and system based on multi-model collaboration, and the method comprises the steps: carrying out the preprocessing of multi-dimensional detection data of a target region, and obtaining a detection image and attached metadata; inputting the detection image and the attached metadata into a lightweight target detection model to obtain a target area image and structured metadata; respectively inputting the detection image, the attached metadata, the target area image and the structured metadata into a multi-modal large model to obtain global semantic information and local semantic information; performing keyword retrieval, vector matching and cross-modal adapter retrieval on the global semantic information in an external professional knowledge base to obtain safety knowledge entries; and inputting the local semantic information, the global semantic information and the safety knowledge entries into a multi-modal large model to obtain a potential safety hazard region, potential safety hazard contents, improvement suggestions and safety regulations. The method is high in positioning precision and high in output reliability, and has professional field knowledge guarantee.

The application discloses a kind of water resource investigation data real-time assimilation methods based on Internet of Things and kilometer grid, comprising: pre-processing to multi-source monitoring data, form kilometer grid data;Quality control is carried out in edge computing node, and hydrological response unit is divided, and state variable is quickly predicted using lightweight machine learning agent model;With DHRU as unit, real-time data assimilation is carried out on edge side;Based on event discrimination, significant change information is uploaded to cloud computing center;Cloud gathers global data and carries out periodic deep assimilation and model parameter global calibration, generates high-precision kilometer grid water resource state product, and the optimized model and parameter are fed back to edge node;Finally, multi-level water resource investigation service product is generated.The application solves the problem that fixed model unit does not match dynamic hydrological process and traditional assimilation calculation efficiency is low, realizes high-precision, high-efficiency real-time sensing and prediction of water resource state.

The invention relates to the technical field of perovskite solar cells, in particular to a perovskite thin film and a preparation method thereof, a perovskite solar cell and a preparation method thereof, and electric equipment. The perovskite thin film comprises an inorganic salt evaporation layer and an organic salt evaporation layer which are sequentially, circularly and alternately stacked. The inorganic salt evaporation layers comprise at least three layers, and the organic salt evaporation layers comprise at least two layers. The thickness of each inorganic salt evaporation layer is 10 to 300nm; and the thickness of each organic salt evaporation layer is 10-400 nm. The general formula of the perovskite thin film is FA1-y-zCsyMAzPbI3-xBrx, 0 < = x < = 1.5, 0 < = y < = 0.5, 0 < = z < = 0.5, and 0 < = y + z < = 0.5. The perovskite thin film not only can realize complete conversion of perovskite, but also can realize optimal matching of the energy level of a perovskite interface, and a high-efficiency perovskite thin film and a perovskite solar cell are obtained.