128results about How to "Improve detection rate" patented technology

The invention discloses an AI-driven abnormal transaction intelligent monitoring and analysis system and method, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining structured features and unstructured features, carrying out the feature fusion, and constructing an account feature vector; constructing a cross-domain heterogeneous graph, learning an embedded vector of an account node in the cross-domain heterogeneous graph based on a time sequence heterogeneous graph neural network, and generating an account portrait vector; training an abnormal transaction detection model, and predicting the abnormal probability of the current real-time transaction record of the account by taking the account portrait vector and the transaction record as input data; dividing all accounts into account groups based on the account portrait vectors, calculating a group abnormal transaction threshold value, and calculating an adaptive abnormal transaction threshold value based on the group abnormal transaction threshold value and the similarity between the account portrait vectors and the center vectors of the account groups to which the account portrait vectors belong; and if the abnormal probability is greater than a self-adaptive abnormal transaction threshold, marking the transaction of the account as an abnormal transaction, triggering risk control measures, and regularly updating the self-adaptive abnormal transaction threshold.

The invention provides an unmanned aerial vehicle group spectrum resource allocation method and device based on cooperative spectrum sensing, and belongs to the technical field of wireless communication, and the method comprises the steps: constructing an unmanned aerial vehicle group dynamic spectrum access network, and creating a cooperative spectrum sensing unit; creating a federal deep reinforcement learning environment, and initializing task parameters and a reward function; the dynamic spectrum access user performs cooperative spectrum sensing to obtain available spectrum resources; training an unmanned aerial vehicle group dynamic spectrum access strategy model based on federated learning and deep reinforcement learning; and calling the trained strategy model to complete dynamic spectrum access of the unmanned aerial vehicle group. Dynamic spectrum access of the unmanned aerial vehicle group is carried out based on cooperative spectrum sensing, efficient utilization of spectrum resources among the unmanned aerial vehicle groups is realized, the problem of shortage of the spectrum resources of the unmanned aerial vehicle group is relieved, the channel utilization efficiency is improved, the interference level is reduced, and conversion of spectrum access of the unmanned aerial vehicle group to intelligence and precision is promoted.

The invention belongs to the technical field of network security, and discloses an Internet of Things attack detection method based on deep learning, which eliminates redundant features through feature selection, enhances inter-class difference and intra-class similarity of selected feature subsets, solves the problem of class imbalance in an attack detection data set by utilizing EQLv2, and improves the attack detection efficiency. And meanwhile, constructing a space-time model to extract space-time information of the sample. According to the method, multi-classification performance evaluation is carried out on an Internet of Things attack detection reference data set NSL-KDD and a CIC-IDS-2017, the advantages of the method are verified from five indexes including the accuracy rate, the precision rate, the recall rate, the F1 score and the false alarm rate, and the effectiveness and feasibility of the method are reflected when the method is compared with a plurality of methods.

The invention provides an in-vehicle living body detection method and system. The in-vehicle living body detection method comprises the steps that continuous multi-frame radar echo data in a vehicle is collected based on a UWB radar installed on the vehicle; for each frame of radar echo data, intercepting radar echo data of each distance unit from the radar echo data; wherein each distance unit is determined according to the vehicle information of the vehicle and the radar information of the UWB radar; extracting the phase of the radar echo data of each distance unit from the radar echo data of each distance unit, and determining the frequency spectrum information of the radar echo data of each distance unit according to the phase of the radar echo data of each distance unit; and determining whether a living body exists in the vehicle or not according to the frequency spectrum information of the radar echo data of each distance unit.

The invention provides a pedestrian re-identification method based on cross-time-domain data, and relates to the technical field of pedestrian identification, and the method comprises the following steps: constructing a monitoring scene; segmenting a monitoring area and a monitoring scene; intercepting a current monitoring video frame to obtain a monitoring image; intercepting an image in the monitoring scene to obtain a background image; a background area in the monitoring image is cut out, and a to-be-processed image is obtained; segmenting the to-be-processed image to obtain an edge image and a center image; capturing a human body target from the edge image to obtain a human body image; giving a unique identity code to the human body target; recognizing the human body image, and marking a recognition result, a correlation mark and an identity code; capturing facial features from the central image to obtain a facial image; acquiring a mark of the human body target; if not, giving a unique identity code to the human body target, recognizing the face image, and obtaining and marking a recognition result; and obtaining, identifying the facial image according to the mark, and obtaining and updating an identification result. And the omission ratio of pedestrians is reduced.

The invention discloses a fault determination method and device, equipment and a computer storage medium. The method comprises the steps of obtaining operation data of multiple devices in a preset time period, constructing a first time-delay matrix and multiple second time-delay matrixes of the operation data by using a sliding window, calculating a first Cauchy-Schwarz divergence of the first time-delay matrix and a target second time-delay matrix, the target second time-delay matrix being any one of the multiple second time-delay matrixes, and calculating a second Cauchy-Schwarz divergence of the target second time-delay matrix; and if the first Cauchy-Schwarz divergence is greater than a first threshold value, determining that the operation data corresponding to the target second time-delay matrix is abnormal, and determining that a plurality of devices fail. In this way, the Cauchy-Schwarz divergence of the first time-delay matrix and the second time-delay matrix is calculated, that is, the fault problem is detected according to the difference degree of probability distribution between the data in the normal state and the test data, the fault is recognized earlier and more accurately, and the detection rate of the tiny fault is increased.

The invention discloses an MMLV reverse transcriptase mutant, which belongs to the technical field of biology and is obtained by mutation design of specific sites of wild type MMLV reverse transcriptase. The mutation site comprises one or a combination of more of A32, S56, E69, L139, T163, T197, H204, S232, N249, T306, F309, T330, G424, L435, D449, D524, E562, D583, L603, H642 and T664, and the mutation site comprises one or a combination of more of the A32, the S56, the E69, the L139, the T163, the T197, the H204, the S232, the N249, the T306, the F309, the T330, the G424, the L435, the D449, the Compared with the existing MMLV reverse transcriptase, the MMLV reverse transcription mutant disclosed by the invention has higher reverse transcription temperature and rate. The invention also discloses a preparation method of the MMLV reverse transcriptase mutant and application of the MMLV reverse transcriptase mutant in reverse transcription reaction.

The invention relates to the technical field of network information security monitoring, in particular to a DPU-based enterprise intranet illegal external connection detection system, which comprises an access control unit, a host monitoring unit, an identity authentication unit, a security management unit, an intranet resource information management unit, a database unit, an automatic detection unit, an alarm unit and a blocking unit. In the invention, the network access examination and approval module registers and examines the physical address of a network card for each host newly accessed into the intranet of an enterprise, the illegal access detection module performs illegal detection on host equipment subjected to network access registration, and the illegal access blocking module can block the communication between the host and other hosts in the intranet. And meanwhile, the automatic detection unit realizes automatic detection of system illegal external connection information through an illegal external connection information automatic detection model, and judges whether the illegal external connection information exists or not, so that illegal external connection of an enterprise intranet of unknown equipment and leakage of sensitive information in an enterprise intranet database system are avoided, and the overall network security of the enterprise intranet is effectively improved.

The invention discloses a multi-feature fusion sea surface small target detection method based on an attention mechanism, and the method comprises the steps: 1, carrying out the preprocessing of radar echo data, and obtaining a combined feature sequence containing rich features; 2, constructing a multi-feature fusion network model based on an attention mechanism, which comprises a feature extraction module, an attention mechanism module, a feature aggregation module and a target detection module, and processing the joint feature sequence to obtain a binary probability matrix sequence; and 3, calculating loss by using a focus loss function, and training the network model to obtain an optimal target detection model. According to the method, through multi-scale feature extraction, an attention mechanism and feature fusion, the detection precision of the radar weak and small target is effectively improved, and the method is suitable for target detection requirements in a complex background and a low signal-to-noise ratio environment.

The invention discloses a millimeter wave traffic radar weak target DBF enhancement detection method. The method comprises the following steps: determining the number of DBF times, the angle of each DBF and the angle range independently covered by DBF beams corresponding to different DBF angles according to a to-be-enhanced detection area and DBF beam width; acquiring detection matrixes corresponding to different DBF beams from the adc data corresponding to each DBF channel according to the detection area needing to be enhanced, the number of DBF times and the angle of each DBF; performing target detection and screening on the detection matrixes corresponding to the different DBF beams according to angle ranges independently covered by the DBF beams corresponding to the different DBF angles; carrying out target detection on other detection areas except the detection area needing to be enhanced through other detection methods; and carrying out combination and parameter estimation on the targets detected by the detection matrix corresponding to each DBF beam and the targets detected by other detection algorithms except for the detection area needing to be enhanced. According to the method, the detection probability of a target with relatively weak echo signals can be obviously improved.

The invention relates to the technical field of radars, in particular to a multi-channel adaptive moving target display and detection synthesis method of a radar. The method comprises a clutter map processing channel: processing a pulse pressure result of an echo by using the clutter map processing channel to obtain a detection result after clutter map cancellation; the ground clutter filtering processing channel is used for processing the echo pulse pressure result to obtain a detection result after ground clutter filtering; the moving clutter filtering processing channel is used for processing the result of the ground clutter filter to obtain a detection result after moving clutter filtering; and multi-channel detection result synthesis: according to the clutter distribution condition, automatically synthesizing and outputting the detection results of the plurality of processing channels. According to the method, the detection probability of the moving target in the ground clutter and moving clutter superposition background can be improved.

The invention belongs to the technical field of flame detection, and particularly discloses a pulse type flame photometric detection device and method.The pulse type flame photometric detection device comprises a detector body, a combustion chamber, an ignition chamber, a combustion chamber air inlet pipe, an ignition chamber air inlet pipe, a light transmission assembly and a signal processing assembly, and an igniter is installed on the ignition chamber; a circulation cavity is formed in the detector, the combustion chamber is installed in the circulation cavity, the ignition chamber is located above the combustion chamber, and the ignition chamber is communicated with the combustion chamber; the ignition chamber air inlet pipe is communicated with the ignition chamber, the combustion chamber air inlet pipe is communicated with the combustion chamber, and the combustion chamber is further communicated with a sample inlet pipe for introducing a measured substance; one end of the light transmission assembly is located in the circulation cavity and located on one side of the combustion chamber, a reflecting mirror is arranged on the right opposite face of the light transmission assembly, the reflecting mirror and the light transmission assembly are located on the two sides of the combustion chamber, the reflecting mirror is installed on the detector body, and the signal processing assembly is used for collecting and receiving photons transmitted in the light transmission assembly. Noise photons can be reduced, and sensitivity can be improved.

The utility model relates to a soil sampler which comprises a movable base, a sampling through groove is formed in the upper end face of the movable base, a stand column is fixedly connected to the upper end of the movable base, a limiting through groove is formed in the surface of the stand column, a threaded sliding plate is slidably connected into the limiting through groove, and the threaded sliding plate is connected with the stand column through a lifting mechanism. A lifting plate slidably connected with the surface of the stand column is fixedly connected to the front end of the threaded sliding plate, a sampling mechanism is arranged below the lifting plate, a rotating disc is arranged on the upper end face of the movable base, and a rotating column rotationally connected with the movable base through a bearing is fixedly connected to the middle of the lower end face of the rotating disc; a plurality of placing circular grooves which are circumferentially and uniformly distributed are formed in the upper end surface of the rotating disc, and sample collecting cylinders are slidably connected into the placing circular grooves. According to the soil sampler disclosed by the utility model, a certain soil sample can be directly taken out for detection treatment according to requirements after various kinds of soil are sampled, so that the soil sampler is relatively practical.

The invention provides an apolipoprotein B detection kit and a use method thereof, the apolipoprotein B detection kit comprises a reagent R1 and a reagent R2, the reagent R1 comprises the following components: 40-80 mmol / L of an HEPES buffer solution, and 1-5 g / L of polyethylene glycol 6000; naCl is 5 g / L to 9 g / L; the preservative is 1g / L to 5g / L, and the surfactant is 1g / L to 5g / L; the pH value of the reagent R1 is 7.0 to 7.5; the pH value is 7.0-7.5 in the reagent R1, so that the pH value is below the isoelectric point of apolipoprotein B, and after the apolipoprotein B and the reagent R1 are mixed, protein in a detection sample can present weak negative charge property; accordingly, hyaluronic acid is added into the reagent R2, the hyaluronic acid and the anti-human APOB antibody in the reagent can form an integrated substance with weak covalent bond combination, and the hyaluronic acid has positive charges, so that apolipoprotein B can be rapidly combined with the anti-human APOB antibody through the attraction of the charges, the problem of long-time incubation is avoided, and the detection accuracy is improved. The detection time is shortened and the accuracy of the detection result is ensured.

The invention relates to the technical field of video analysis, and discloses an algorithm for judging vehicle illegal parking through vehicle thermodynamic diagram statistics, which sequentially comprises the steps of video frame extraction, road area recognition, vehicle recognition, thermodynamic area acquisition, candidate target confirmation and illegal parking judgment. According to the method and the device, after the vehicle target is identified, the strategy of tracking the vehicle is not directly adopted, the thermodynamic diagram of a period of time is counted, the area with the high hot spot is obtained through the thermodynamic diagram statistics, then vehicle target matching is carried out on the area with the high hot spot so as to obtain the suspicious candidate target, and then template matching is carried out on the candidate target for a period of time. And judging whether the position is moved so as to obtain a conclusion whether the vehicle is illegally parked or not. Tracking targets can be effectively reduced through thermodynamic diagram statistics, the algorithm calculation amount can be reduced, then target movement judgment is carried out through an image matching algorithm instead of a tracking algorithm, and the detection rate is improved.

The invention discloses a method for optimally extracting azimuth information of a weak target, and belongs to the field of underwater acoustic equipment design and manufacturing. The method comprises the steps of firstly specifying an application scene and making a reasonable hypothesis, then deducing the change of a target orientation, then performing orientation prediction, then predicting the target orientation, and finally verifying simulation data. According to the invention, after the signal received by the passive sonar is subjected to beam forming and other signal processing, the azimuth history diagram of the target is formed. In order to obtain the azimuth information of the target, the azimuth history chart is processed to extract the target azimuth. A conventional target orientation extraction method is a peak value detection method, but processing of a weak target is omitted outside a detection threshold due to weak energy of the target. The objective of the invention is to optimize the peak detection method, predict the azimuth when the temporary detection of the target azimuth disappears, and improve the detection rate of the target azimuth.

The application discloses a threshold segmentation method for an infrared small target background suppressed image, which realizes threshold segmentation of the infrared small target image after background suppression by calculating the similarity between a Gaussian filter template and an image block, fully utilizes the gray scale features and structural features of the small target, and accurately extracts all the small targets in the image. The application improves the problem that the traditional threshold segmentation method is prone to missing small targets in a multi-target detection task, realizes image segmentation with high detection rate and high real-time rate, provides guarantee for accurate detection of the infrared small target, and has the advantages of simple segmentation process, high efficiency and the like.

The application relates to a wireless communication same-frequency interference detection method based on a CNN-SVM and a particle swarm optimization, which comprises the following steps: performing short-time Fourier transform on a received baseband signal to generate a log power spectrum time-frequency graph; automatically extracting a deep space feature vector through a convolutional neural network; performing global self-adaptive optimization on a penalty parameter and a kernel parameter of a support vector machine through a particle swarm optimization algorithm; and training the support vector machine with optimal parameters to perform binary classification judgment on a to-be-detected feature. Through a triple coordination mechanism, the application overcomes the problems of fixed parameters, poor robustness and the need for a large amount of labeled data in traditional methods, and realizes high detection probability and low false alarm rate under different signal-to-interference-and-noise ratios and bandwidth conditions.

The invention discloses a method for detecting specific pollutants in a pesticide production plot, which can simultaneously detect six specific pollutants including three organophosphorus pesticides including isocarbophos, isofenphos-methyl and profenofos. One pesticide degradation product is dimethyl phosphate; the intermediate product of one pesticide is dimethyl phosphite, and the production raw material of the other pesticide is dimethyl sulfate. The method comprises the following steps: carrying out freeze-drying and grinding on a soil sample, and sieving with a sieve with the aperture of 0.25 mm; uniformly stirring the soil sample and diatomite, and adding a substitute and an extracting agent for extraction treatment to obtain a sample extracting solution; carrying out dehydration treatment on the sample extracting solution; and performing concentration treatment on the sample extracting solution to obtain a to-be-detected sample solution containing six target objects. The method can be used for simultaneously detecting specific pollutants of 6 pesticide production plots, and the detection rate is high; by adding an interior label, the accuracy of the method is improved; and the qualification rate of sample detection is improved by adding a substitute.

The present invention discloses a method for detecting pathogens by combining a SERS-based CRISPR / Cas system with a microfluidic paper analysis device and its application. The method only requires extracting the genome of the pathogen in a sample, and after RPA isothermal amplification, the CRISPR / Cas system can be used to react with a sandwich gold nanostar probe on a μPAD to achieve detection of the microbial pathogen Salmonella. The detection range is wide, the method is simple, fast and easy to operate.

The invention relates to the technical field of detection equipment, and particularly discloses an automatic detection device and a detection method thereof.The automatic detection device comprises a rack and a workbench arranged on the rack, the workbench is slidably connected with multiple sets of bearing tables, and each bearing table is provided with multiple sets of insulin needle cylinder workpieces through an extrusion assembly; a detection mechanism is arranged on the machine frame and used for detecting the insulin needle cylinder workpiece, a trigger assembly is arranged between the machine frame and the working table and located on one side of the detection mechanism, and a driving assembly is arranged on the working table and used for driving the bearing tables to slide synchronously. The driving part drives the bearing table to slide on the workbench, the extrusion assembly is driven by the triggering assembly to fix the multiple sets of insulin needle cylinder workpieces installed on the bearing table in the sliding stroke, and after the insulin needle cylinder workpieces are fixed, the insulin needle cylinder workpieces are detected through the detection mechanism.

The present invention relates to the field of Internet information processing technology, and specifically to a sensitive information detection method based on a deep model and a knowledge base, S1. Preprocessing text information input by a user; S2. Detecting sensitive information on the text information through a BERT deep model and a knowledge base; S3. Filtering the sensitive information detection results; S4. Merging and outputting the filtered sensitive information. The present invention can improve the accuracy of sensitive information detection.

The invention discloses a carbon nanotube conductive paste solid content testing device, which comprises a testing workbench, a sample introduction mechanism, a heating mechanism, a centrifugal rotation mechanism and a weighing mechanism, the sample introduction mechanism is arranged on the testing workbench, the sample introduction mechanism is used for loading a carbon nanotube conductive paste test sample, the heating mechanism is arranged on the inner side of the bottom of the sample introduction mechanism, and the centrifugal rotation mechanism is used for loading the carbon nanotube conductive paste test sample. The heating mechanism is used for heating and evaporating a carbon nanotube conductive slurry test sample in the sample introduction mechanism, the weighing mechanism is arranged below the sample introduction mechanism and is used for measuring weight data of the carbon nanotube conductive slurry test sample, the centrifugal rotating mechanism is arranged in the sample introduction mechanism and comprises a rotating seat, and the rotating seat is arranged on the rotating seat. The upper portion of the rotating base is of an opening structure, and filter holes are formed in the side edge of the rotating base and provided with filter membranes. The method has the beneficial effects that the test time can be remarkably shortened, and meanwhile, the test accuracy is improved.

The invention relates to a full-blind multi-master-user spectrum sensing method based on extreme value characteristic values, which is characterized in that a sensing judgment quantity is constructed by using a difference value of the extreme value characteristic values and harmonic average: firstly, a multi-antenna sensing node calculates a sampling covariance matrix # imgabs0 # by using a received signal vector so as to calculate a maximum characteristic value and a minimum characteristic value of the sampling covariance matrix # imgabs0 #; thirdly, constructing a perception decision amount # imgabs1 # by using the ratio of the difference value of the extreme value characteristic values to the harmonic average value, and calculating a decision threshold t based on the analysis result of the extreme value characteristic values; and finally, carrying out perception judgment: when # imgabs2 is greater than t, judging that a multi-master user signal exists; otherwise, determining that the multi-master user signal does not exist. The method provided by the invention has reliable false alarm performance, and the detection probability higher than that of classical maximum and minimum characteristic value ratio and covariance absolute value detection can be obtained under the condition of low signal-to-noise ratio; and the method can realize total blind detection, the judgment threshold can be directly calculated based on closed expression, and the method has the advantages of simplicity and convenience in calculation and low complexity.

The invention relates to a DPU-based fog cloud collaborative Internet of Things intrusion detection system and method. The system comprises an edge node used for receiving target data sent by Internet of Things equipment, encoding the target data through a pre-trained stacked auto-encoder to obtain a plurality of encoding features, determining a prediction capability index of each encoding feature, and determining a target encoding feature according to each prediction capability index; the cloud server is used for receiving the target coding features sent by the edge node, inputting the target coding features into a pre-trained Internet of Things intrusion identification model, extracting target global features, target local features and target time features associated with the target coding features and Internet of Things intrusion through different branch networks in the Internet of Things intrusion identification model, and sending the extracted target global features, target local features and target time features to the edge node; and outputting an Internet of Things intrusion detection result according to the target global feature, the target local feature and the target time feature. By adopting the system, Internet of Things intrusion detection with low resource occupation, high precision and high efficiency can be realized.

The utility model discloses a surface smoothness detection device, which belongs to the technical field of thrust rod ball seat detection, and comprises a bottom plate and a workbench located at the top of the bottom plate, a motor is installed at the top of the bottom plate, a rectangular sliding rod is vertically installed at the output end of the motor, a supporting rod is vertically arranged at the top of the rectangular sliding rod in a sliding mode, and a rectangular sliding hole is formed in the bottom of the supporting rod. The rectangular sliding rod is located in the rectangular sliding hole, and a lifting mechanism is arranged on the outer side of the supporting rod. The lifting mechanism composed of the transverse plate, the vertical rod, the cylindrical rod, the rotating rod, the check ring, the inclined plate and the passing opening is arranged on the outer side of the supporting rod, automatic rotation of the rotating rod is controlled in a gear transmission mode, and in the process of controlling rotation of the thrust rod ball seat, height adjustment can be automatically conducted after the thrust rod ball seat rotates by a circle every time; the surface of the thrust rod ball seat is uniformly detected, manual adjustment is not needed, use is more convenient, and the detection speed is increased.

This invention discloses an air conditioning device that determines the initial and equilibrium temperatures of a power module; controls the compressor's operating state to ensure the IGBT output current has a relative zero temperature coefficient; determines the total power module losses at the equilibrium temperature based on the IGBT's relative zero temperature coefficient parameters; acquires the thermal resistance of the inverter driver and refrigerant radiator, and determines the theoretical temperature based on the thermal resistance and total losses; determines the theoretical temperature rise based on the theoretical and initial temperatures, and determines the measured temperature rise based on the equilibrium and initial temperatures; determines the deviation value based on the theoretical and measured temperature rises; acquires the standard deviation value, and compares the deviation value with the standard deviation value to determine whether the refrigerant radiator is qualified. The air conditioning device can intelligently assess the reliability of the refrigerant radiator and its installation.

The application provides a large frequency offset processing method suitable for a multi-user scenario of PUCCH format 1, which comprises the following steps: obtaining user-level PUCCH format 1 frequency domain data, performing zero setting on the first and last subcarriers; performing IFFT on the subcarrier dimension to obtain time domain data from which adjacent user interference is removed; calculating a frequency offset compensation factor according to a historical frequency offset and a sampling rate conversion relationship; multiplying the time domain data and the frequency offset compensation factor to obtain frequency offset compensated time domain data; obtaining frequency offset compensated frequency domain data; taking out PUCCH data and pilot symbols, multiplying the data and the pilot symbols with a data and pilot template to obtain LS channel estimation results; and filling the zero values at the positions of the first and last subcarriers of the LS with the LS channel estimation results of adjacent subcarriers. The method of the application performs zero setting on the first and last subcarriers of the user-level frequency domain data, reduces the interference caused by power leakage of other frequency division adjacent users, and is suitable for a frequency division multi-user large frequency offset communication scenario.

The present invention belongs to the field of infrared information acquisition and image processing technology, and discloses an infrared line array camera point target search system, control method and application. The output end of a one-dimensional parallel search and clutter suppression device is connected to a morphological operation device, and the output end of the morphological operation device is connected to a point target detection device; the one-dimensional parallel search and clutter suppression device is provided with an imaging data distributor, a parallel data processor and a parameter corrector; the clutter suppression device is provided with a target expansion operator; the point target detection device is provided with a memory, an operator and a judger. Based on the imaging characteristics of the infrared line array camera, the present invention first performs a one-dimensional parallel rapid search to reduce the influence of the non-uniformity of the infrared camera, mark the potential target point, and then perform a morphological operation on the target point. Finally, according to the preset parameters, the temporal and spatial correlation between frames is used to confirm the target. The use of the present invention can significantly improve the signal-to-noise ratio of point targets in infrared remote sensing images and improve the detection rate.

The invention discloses a target detection method and system based on sparse region extraction. The method comprises the following steps: firstly, extracting features of a to-be-detected image and performing fusion enhancement to obtain enhanced features with different resolutions; secondly, for high-resolution features, screening out a feature window through a sparse region extraction module; a first prior frame is generated based on the feature window, and the category and offset of the first prior frame are predicted through a first target detection head module; meanwhile, a second prior frame is generated for other enhanced features, and prediction is carried out through a second target detection head module; and finally, adding all prior frames into the candidate anchor frame set after offset according to the offset, and obtaining a final target detection frame through non-maximum suppression. According to the method, sparse region extraction is carried out on high-resolution features, key partial regions are screened to carry out small-size target detection, and normal global target detection is carried out on other low-resolution features, so that the detection efficiency is improved while the detection accuracy of the small-size target is improved.