45results about How to "Suppress background distractions" patented technology

The invention provides an object significance detecting method based on color contrast and color distribution. The steps of the object significance detecting method based on the color contrast and the color distribution include that S1: an input image is divided into small size super-pixels, average color and position in the super-pixels are calculated; S2: the center-periphery color contrast of each super-pixel is calculated, the color contrast value is multiplied by a priori distribution, and at last color contrast significance diagram is obtained by using a significance smooth operation; S3: color distribution variance of each super-pixel is calculated, and thereby a color distribution significance diagram is obtained; S4: the color distribution significance diagrams obtained by the S2 and the S3 are multiplied and refined by using MeanShift division, edges of an object are enabled to be more fine, and the final significance diagram is output. According to the object significance detecting method based on the color contrast and the color distribution, the significance diagram obtained can evenly highlight the significant object in the significance diagram, the edge details of the object are well retained, the background interference is restrained, and the following up processes such as the target object division are benefited.

The embodiment of the invention relates to the electric power inspection technical field, and discloses a power transmission line detection method and device; the method comprises the following steps: carrying out frame difference operation between two adjacent frame images of a collected visible light video sequence so as to obtain a difference image, and binarizing same to obtain a binary difference image; using a preset threshold edge detection algorithm to carry out edge detection for the binary difference image so as to obtain an edge binary difference image; carrying out Hough conversion so as to detect linear sections; carrying out cycle iterations and screening merging for detected linear sections according to the transmission line parallel characteristics until the obtained target linear section number is smaller than a preset value, and marking and displaying same; calculating the central position of a target zone, and sending the central position to a hang cabin so as to adjust the posture of the hang cabin. The method and device can effectively improve the power transmission line detection efficiency and accuracy, thus reducing work load of inspection personnel.

The invention provides a detection method for salient regions of images based on geodesic line distance. The method comprises the steps of S1, over-dividing input images into super-pixels of small areas, and calculating average colors and positions in the super-pixels; S2, trasversing all the super-pixels to obtain rough saliency energy values; S3, detecting and sequencing Harris angular points in the images, reserving a certain number of angular points with high scores, and searching minimum circumscribed convex hulls of the angular points; S4, setting the rough saliency energy values of the super-pixels outside the convex hulls to be 0, and keeping the saliency energy values of the super-pixels in the convex hulls to be unchanged; and S5, calculating final saliency values of the super-pixels in the convex hulls according to spreading saliency energy of geodesic line distance. According to the detection method for salient regions of images based on geodesic line distance, saliency objects or areas in the images can be uniformly highlighted; and simultaneously, background interference can be strongly restrained, and subsequent processing such as target object dividing is facilitated.

The invention discloses a multiscale infrared target adaptive detection method based on the visual reception field, relates to a multiscale infrared target adaptive detection method, and belongs to the technical field of machine vision. A calculation differential operator of an edge detector is used to calculate the gradient direction theta of an image, and an orientation information distribution map of pixel points is obtained; a local background is predicted according to background information similar to local characteristic in the infrared image; a mathematical model Gabor filter in the simple cell reception field is used to filter an object image, an object is detected preliminarily, and a direction parameter theta of the Gabor filter is determined adaptively by orientation information distribution of the pixel points; and the gray scales of the pixel points are adjusted to obtain a background inhibited and object enhanced image, and the detection precision is improved. According to the invention, the robustness, anti-interference capability and adaptability, to different scales of object, of the infrared object detection method can be improved, and the operand is low.

The present invention relates to a double-waveband infrared flame detector and its detection method. Said detector includes supporting frame, external shell, protective lens, filter, infrared sensor, amplification circuit, signal processing circuit, power supply conversion circuit, signal output circuit, internal and external shielding covers,detector interface plate and setting circuit. Said invention also provides its working principle and concrete steps of its detection method.

The invention discloses an infrared image dividing method and system for power system equipment based on wavelet analysis. Firstly, wavelet conversion, fuzzy entropy, a genetic algorithm and a mathematical morphology are used for carrying out image processing on an infrared image of the power system equipment. The method comprises the following steps: firstly, carrying out thermal failure detection on the power system equipment by an infrared thermal imager to obtain a thermal image; eliminating mixed noise of the thermal image by using the wavelet conversion to inhibit background interferences and enhance a target; carrying out combined optimization operation by applying the fuzzy entropy and the genetic algorithm to determine an optimal threshold value; extracting a target; solving a discontinuous boundary problem by using a waterline region dividing method of the mathematical morphology and dividing the image so as to find a largest communication region; and separating a target region. Finally, a position of a failure point of the power equipment and an element with a failure can be judged clearly according to the separated target region; an accident is prevented form occurring and overhauling under poweroff is not blindly carried out; the operation reliability of a power system is improved.

The present invention relates to a dual-channel infrared behavior identification method based on attitude profile and motion information. The method comprises the following steps of: S1, in a motion information channel, extracting variation optical flow characteristics of an original infrared image sequence, and obtaining an optical flow image sequence; S2: dividing the optical flow image sequenceinto a training set and a test set, and after fine tuning of a C3D network, extracting network characteristics; S3: in an attitude profile channel, performing attribute estimation of an infrared image, and extracting a human body skeleton diagram; S4: performing Gaussian smoothing of the human body skeleton diagram, and obtaining an attitude characteristic graph; S5: dividing the attitude characteristic graph into training test sets, and after fine tuning of the C3D network, extracting network characteristics; S6: performing early-stage fusion of the characteristics of the two networks, and obtaining final characteristics; and S7: employing an SVM classifier to perform classification identification of the final characteristics to obtain an identification result. The dual-channel infraredbehavior identification method based on attitude profile and motion information comprehensively employs video information to take the attitude characteristic graph as more effective profile information on the space and improve the identification result so as to provide a technical support for subsequent research.

The invention belongs to the technical field of computer vision and discloses a method for detecting the contour of a target object. The method comprises the following steps of: dividing a non-classical receptive field into an end area and a side area; performing constant inhibition in the side area; performing selective action in the end area by using influence of contrast; during computation of the selective action, setting a threshold according to a contrast value in a certain neighborhood of each pixel by using an adaptive method to judge whether the pixel is inhibited or facilitated; according to an inhibitory action of the side area of a non-classical receptive field filter on the pixel at the center position, suppressing texture and other background information with similar local characteristics, and further highlighting contour information; and according to the inhibitory action of the side area of the non-classical receptive field filter, regulating the size of the action of the end area to preserve the completeness of a contour line as better as possible and inhibit background interference to the largest content. The method for detecting the contour of the target object can be applied to detection, identification and the like of the target object in the computer vision.

The invention relates to a multi-light path optical fiber fluorescent sensor. In the sensor, two or three excitation light sources are adopted; each light source is correspondingly coupled with one transmitting optical fiber respectively; the receiving end of a receiving optical fiber is combined with the output end of an excitation optical fiber in a pair and an angle of between 15 and 45 degrees is formed; and fluorescent light is combined together and enters a photovoltaic conversion device to be detected through the same filter after being received and transmitted by two or three receiving optical fibers. Multi-light path design is adopted in the sensor, and measuring ends are a given distance away from each other, so that the accuracy and the repeatability for measuring the concentration of a fluorescent substance in a non-uniform fluorescent medium can be improved, and the irradiation region of exciting light is also improved by 1 to 1.5 times to further improve the intensity of the fluorescent light and the flexibility of the detection. The sensor has a simple structure and is convenient to use and low in cost, non-solid fluorescent substances such as gas, liquid and the like can be directly detected, and the sensor can be combined with a fluorescent sensing film to detect non-fluorescent substances.

The present invention discloses a camera module optical filter gluing detection apparatus and method. The method includes the following steps: aiming at the morphology and the reflection characteristics of the glue smeared on the surface of the camera module optical filter, using a low-angle ring illuminator to obtain a clear picture of the glue; using a dynamic threshold algorithm to extract a glue target; then using morphological operations for preprocessing and conducting communication operations to divide areas; and finally extracting the basic parameters of the glue, using a support vector machine for classified judgment of the normality, and if abnormity exists, providing the abnormity type. The method can ensure high-speed gluing test in an open state. The method uses a non-contact detection method with high real-time performance, can effectively detect various types of gluing problems, and has recognition accuracy rate of higher than 99%.

The invention discloses an infrared filter with a broadband of 3.65 to 5 microns, and a manufacturing method for the infrared filter. The infrared filter with the broadband of 3.65 to 5 microns is characterized in that: monocrystalline germanium Ge with the size of phi 18*1.0mm is used as a base plate 1, wherein the surface aperture N of the monocrystalline germanium Ge is less than or equal to 5, and the local aperture delta N is less than or equal to 0.5, the depth of parallelism theta is less than or equal to 1', and the surface finish is more than 60 / 40; coating materials are silicon monoxide SiO and the monocrystalline germanium Ge, and multiple layers of interference films are deposited on two surfaces of the base plate; Sub 1.33(.5HL.5H)<5>3.39(.5LH.5L)<5> Air is adopted by the design of a first surface film system interference film 2; Sub 0.92(.5HL.5H)<3> 1.1(.5HL.5L)<4> Air is adopted by the design of a second surface film system interference film 3; and according to the infrared filter with the broadband of 3.65 to 5 microns, gradient of two sides is less than 3 percent, namely the wavelength of a transition area from the transmissivity of 5 percent to the transmissivity of 80 percent is less than 20nm, the transmissivity of a high-transmissivity area is more than or equal to 90 percent, the cut-off depth in a cut-off area is less than or equal to 0.1 percent, the transmissivity reaches more than 92 percent in the range of thermal imager wavelength of 3.7 to 4.9mu m, excellent signal-to-noise ratio is achieved, and the requirements of sensitivity and accuracy of a high-performance thermal imager can be met.

The invention discloses a moving vehicle detection method based on aerial video of an unmanned aerial vehicle, at first, the SURF algorithm is used to match the feature points of the image and removethe anomaly point, a UAV image registration algorithm combining global and local homography matrices is used to obtain the transformation matrix, compensating for the adverse effects of airborne camera motion, a 2-frame difference method is used to reduce the area to be detected, and then traverses the region to be detected according to the center of the superpixel, then, the low-order features are extracted by using the multi-channel HOG feature algorithm, and the high-order features are obtained by introducing the context information of the vehicle, and the two features are fused to obtain the multi-order features of the target vehicle. Finally, the multi-order features and dictionary learning algorithm are combined to realize the moving vehicle detection. The method can suppress the influence of the motion of the airborne camera of the unmanned aerial vehicle, process the vehicle deformation and background interference in the image, and improve the robustness and real-time performance of the moving vehicle detection.

The invention discloses a weak and small target rapid detection method based on a high-resolution optical remote sensing image, and mainly solves the problems of high missing detection and false detection and low detection speed in the prior art. According to the implementation scheme, the method comprises steps of converting a to-be-detected multispectral four-channel image into an 8-bit RGB color image and then converting into a grey-scale map; performing extended wavelet transform on the grey-scale map; carrying out corner point detection on the image after extended wavelet transform, filtering dense corner points in space, and reserving scattered corner points; carrying out morphological closing operation of expanding and then corroding on the reserved corner points to obtain an area after the morphological closing operation; extracting shape parameters from the area after morphological closing operation, calculating contrast and correlation, combining the three parameters to screen angular points, and a local area where the finally reserved angular points are located being a target area. According to the method, missing detection and false detection are greatly reduced, the detection accuracy and the detection speed are improved, and the method can be used for detecting satellite remote sensing images with backgrounds of mountainous regions, forests and deserts.

The invention discloses a 7.6-micron front cut-off infrared optical filter and a making method thereof. The optical filter is characterized in that: single crystal germanium (Ge) with the size of phi 25.4*1.0mm is used as a substrate, the surface aperture N of the substrate is less than or equal to 5, the local aperture delta N is less than or equal to 0.5, the parallelism theta is less than or equal to 1', and the surface finishment is superior to 60 / 40; zinc sulfide (ZnS) and single crystal germanium (Ge) are used as coating materials, and a plurality of layers of interference films are respectively deposited on two surfaces of the single crystal Ge substrate; Sub absolute value 2.42(0.5HL0.5H)<8>3.0(0.5HL0.5H)<7> Air is adopted in the design of a first side film system; Sub absolute value 1.2(0.5HL0.5H)<6>1.6(0.5HL0.5H)<7>2.32(0.5HL0.5H)<6> Air is adopted in the design of a second side film system; according to the adopted evaporation process conditions, heating and baking of below 300 DEG C are performed in a vacuum environment with the vacuum degree of less than or equal to 10<-3>Pa, and a physical vapor deposition mode and ion source aided coating are adopted; and the 7.6-micron front cut-off infrared optical filter has excellent signal-to-noise ratio and good background interference suppression. The optical filter can be widely used for detecting the temperature of a far infrared object.

The invention discloses a polarized synthetic aperture radar image ship detection method based on a convolutional neural network. The method comprises the steps of S1, carrying out the data preprocessing of a polarized synthetic aperture radar image, and constructing a polarized synthetic aperture radar image ship detection data set representing the polarization features and geometric features; S2, constructing a double-branch convolutional neural network for polarization feature extraction and geometric feature extraction, and constructing a classification regression task sub-network through the polarization features and the geometric features; and S3, training the dual-branch convolutional neural network based on the polarized synthetic aperture radar image ship detection data set according to preset training parameters, a loss function and a training strategy, and executing polarized synthetic aperture radar image ship detection and index evaluation by using the dual-branch convolutional neural network. According to the invention, the identification capability of the network can be greatly improved, and the ship can be effectively detected from a complex near-shore open-sea scene.

The invention provides a heart segmentation model and pathology classification model training, heart segmentation and pathology classification method and device based on heart MRI (Magnetic Resonance Imaging), and the method comprises the steps: suppressing a residual background part with a small pixel gray level change through a standard deviation filter, highlighting a left ventricle, a right ventricle and a myocardial ,the central position of the left ventricular myocardial wall being further obtained through canny edge detection and circular Hough transform, drawing a rectangular mask, the two-dimensional image being cut based on the rectangular mask to serve as input for training a preset neural network model for training. Background interference can be greatly inhibited, and fast convergence of neural network training is promoted. The pathology classification model training method comprises the following steps: segmenting a two-dimensional image obtained by segmenting each frame of cardiac magnetic resonance imaging short axis in a cardiac cycle based on a cardiac segmentation model, calculating classification feature values, and constructing a random forest based on the classification feature values of a plurality of samples and pathology classification to obtain a cardiac pathology classification model; and realizing automatic pathological classification.

The invention discloses a real-time color object tracking method. The method comprises the following steps: selecting a target template under the particle filtering frame, and initializing a particleset; obtaining a current frame, and obtaining a new particle set through a state transfer model; extracting the target color information and movement information, carrying out the drifting of each particle of the new particle set through an improved mean value drift algorithm, and obtaining a particle set after drifting; building a new observation model based on the fusion of the target color information and the movement information, and calculating the particle weight of the particle seat after drifting; carrying out the normalization of the particle weight, and obtaining the state of the current frame through the weighing of the particles. The method provided by the invention can iron out the defect that the particle filtering calculation burden is large, can inhibit the background interference, and achieves the stable and real-time tracking of a target.

The invention provides a stereo matching method based on translation and combined measurement of salient images. The method comprises that salient areas are respectively extracted from a left image and a right image shot by a binocular camera, and according to the salient images, a rough matching method based on translation of the salient images is provided to realize rough matching; and SAD (Sum of Absolute Differences) matching is implemented on the basis of rough matching, structural similarity (SSIM) is used for fine matching at the minimum of SAD, and the position with the highest SSIM score is selected as the best matching point. The stereo matching method can effectively improve the matching precision and reduce the computational complexity.

The invention discloses a medium-wave infrared narrow band filter with a wavelength of 5.25 microns, which is prepared by taking an aluminum oxide crystal as a substrate and taking zinc selenide and lead telluride as film layer materials through a vacuum film deposition method. A main film system is in a multi-half-wave and high-order separating layer film system structure and is matched with a secondary peak cutting-off film system with high cutting-off depth, thus the main film system has good effect on inhibiting interference of other gases. The filter is a critical component of a nitric oxide infrared analyzer and is suitable for batch production. The optical performance of a product and the physical strength of a film layer meet the actual use requirements. The invention is suitable for trace nitric oxide gas analyzers in the fields of environmental protection, petrochemical industry and the like.

The invention discloses a method for extracting rusty area on a fitting surface and belongs to the technical field of electric power automatic detection. The invention aims to provide a method for extracting a rusted area on the surface of a hardware fitting, and solves the problems of poor universality of infrared or ultraviolet detection, easiness in background interference and serious visible light imaging overexposure phenomenon in the prior art, and difficulty in positioning the rusted area. The method comprises the steps of in the process of aerial photography of a hardware fitting surface image by an unmanned aerial vehicle, obtaining polarization images of different polarization angles in cooperation with a linear polaroid, and then obtaining a polarization degree map; constructinga saliency map for the polarization degree map, performing saliency feature filtering, and further enhancing a rusted region in the polarization degree map; and carrying out binarization extraction on the processed image, setting an area threshold value and a length-width ratio threshold value, and extracting a rusted region. According to the method, the problem of non-uniform overexposure of hardware surface imaging under natural light is avoided, the reliability and the detection efficiency of rust area extraction are effectively improved, and technical support is provided for later accurate analysis of the characteristics of the rust area.

The invention provides an induced polarization deep prospecting method comprising: by using a dipolar device induced polarization method, using a current source to supply power to the dipolar device in order to acquire a shallow polarizability [eta]1; and using a voltage source to supply power to the dipolar device, adjusting a ratio of T-time current to 0-time current of a power supply loop in order to enable an equation I(T) / I(0) = 1-[eta]1, eliminating the shallow polarizability [eta]1 and measuring the abnormality of deep induced-polarization body apparent polarizability. The induced polarization deep prospecting method has a deep detecting capability, inhibits shallow mineralization high background interference, and highlights the abnormality of the deep induced-polarization body.

The invention relates to an array light source induced fluorescent optical fiber sensor. In the sensor, N light-emitting tubes are used as excitation light sources (N is more than or equal to 2); each light-emitting tube is respectively coupled with an excitation optical fiber correspondingly; the excitation optical fibers are arranged in the form of surrounding an axis; and a receiving optical fiber is arranged on the central line of the axis. In the sensor, a plurality of light-emitting tubes are adopted as the light sources, which not only can effectively improve the exciting light intensity in a target irradiation area by N times, and improve the light intensity-noise ratio by the square root of N, but also can effectively inhibit the background interference of the external environmental light, improve the signal to noise ratio of the sensor, improve the sensitivity of detection and the external light anti-interference ability, and improve the receiving efficiency of the fluorescence and the sensitivity of detection through optimizing the included angle of the excitation optical fiber and a transmission optical fiber. The sensor is simple in structure, convenient in use and low in cost; and by the sensor, non-solid fluorescent substances, such as gas and liquid and the like can be directly detected, and non-solid fluorescent substances can be detected after being combined with a fluorescent sensing membrane.

The invention provides a quick beamforming method capable of improving array resolution and gain, comprising the following steps: adopting the construction of a minimum redundant array to optimize an M-element uniform linear array into a P-element non-uniform linear array; carrying out FFT processing on primitive data of a P-element array; in a frequency domain, constructing a covariance matrix of data based on the uniform linear array in accordance with the array aperture extension characteristics of fourth-order cumulants; carrying out normalization processing on the beam space and carryingout estimation on a Bartlett spatial spectrum. By adopting the array aperture extension characteristics of the fourth-order cumulants, the invention realizes that the optimized element layout form isemployed to obtain high resolution and overcomes the defects of high requirements of the original fourth-order cumulant-based methods on snapshots and great computational complexity to enable the computation process to be simple and easy to operate. When the signal to noise ratio is higher than the supercritical signal to noise ratio, the method of the invention has higher array gain than that ofconventional beamforming. The normalization processing of the beam space realizes effective inhibition to background interference. The beamforming method of the invention is simple and easy to operate and is especially suitable for project application.

The invention provides a landmine detection system and method capable of suppressing a background. The landmine detection system comprises the following parts: a transmitting coil module and a transmitting circuit module are connected; a receiving coil module is connected with a receiving and amplifying circuit module; a transmitting circuit module and the receiving and amplifying circuit module are connected with a quadrature demodulation module; the quadrature demodulation module is connected with a digital signal processing band-pass filter module; the digital signal processing band-pass filter module is connected with the transmitting circuit module; the method is better in background suppression performance, can extract detection background echo signals of a plurality of frequency points compared with a conventional single frequency, and can form a statistical filter.

The invention discloses a short-wave infrared narrowband optical filter. The short-wave infrared narrowband optical filter is characterized in that the short-wave infrared narrowband optical filter comprises a substrate and a front film system and a reverse film system formed at surfaces of the substrate at two sides, the front film system is A / (HL)<4>L(HL)<8>L(HL)<8>L(HL)<4>1.64H0.64L / S, the reverse film system is A / (0.5HL0.5H)<11>[alpha](0.5HL0.5H)<12>[beta](0.5LH0.5L)<7>[gamma](0.5LH0.5L)<10>[omega](0.5LH0.5L)<10> / S, and in the film systems, A refers to air, S refers to an H-K9L glass substrate, H refers to a high-refractive-index material titanium pentoxide, L refers to a low-refractive-index material silicon dioxide, and alpha, beta, gamma, and omega respectively refer to multiples of the central wavelengths of the film systems to the central wavelength. According to the optical filter, the infrared narrowband optical filter transmission spectrum with the central wavelength of 1240 nm is provided, the rise and the decline of a transmission band are less than 0.5% along the steepness, the passband average transmittance reaches 80%, the stop-band average transmittance is less than 0.5%, the spectrum range is limited, the background interference is suppressed, and the target resolution is improved.

The invention discloses a depth shape priori extraction method. The depth shape priori extraction method comprises the following steps of selecting K super-pixel regions with relatively high RGB significance values as root seed points, and establishing a relationship between a depth characteristic and significance; based on depth smoothness and consistency constraints, determining a sub-node set of each root seed point to describe depth shape attributes; and by considering depth consistency of related super-pixel nodes in two continuous circulation propagation and depth consistency between super-pixels in current circulation and the root seed points, defining a final DSP value as a maximum value of the depth consistency in the two conditions; and fusing DSP images generated by the root seed points to obtain a final DSP result. Through deep analysis on depth image data, shape priori information of a depth image is fully mined, and effective depth information is provided for RGBD significance detection.

The invention discloses an anti-blocking and multi-scale dead fish recognition system and method based on deep learning, and aims to improve a Faster RCNN recognition model for the difficulties of complex dead fish recognition background, large scale change, blocking and the like so as to adapt to a dead fish recognition scene in a complex underwater environment. The method comprises the following steps: carrying out contrast-limited adaptive histogram equalization processing on a to-be-recognized image to enhance the local contrast of a target; designing a multi-scale feature enhancement module to solve the problem of dead fish recognition in a large-scale range, designing an anti-shielding module based on an attention mechanism, highlighting a target area, and eliminating background interference such as other noise; finally, utilizing the rotating rectangular frame to represent the dead fish target, so the recognition precision of the dead fish in the dense scene is greatly improved.

The invention provides a traffic person and vehicle non-target detection method for a small target. The method comprises the following steps: firstly, obtaining a to-be-trained traffic image sample set; performing data enhancement on the data of the small target; then constructing a small target detection network based on the yolov5, inputting the image sample set with the label into a neural network, updating a parameter weight of the neural network according to a partial derivative of a loss value to a network parameter, and when the loss value is minimum, obtaining a trained human-vehicle non-target detection model for a small target; and finally, inputting a to-be-detected traffic picture into the detection model, and outputting the category, confidence and four position coordinates of each target in the image. According to the invention, by using a data enhancement method for small targets and constructing a feature fusion module based on an attention mechanism, real-time detection of non-targets of traffic people and vehicles is realized based on a YOLOV5 detection network, and the detection rate and the detection accuracy of the small targets are greatly improved.

The invention relates to a track fastener image positioning method based on superpixel node sorting. The track fastener image positioning method comprises the following steps: 1) shooting and collecting an original image of a track fastener; 2) carrying out image preprocessing and super-pixel segmentation to obtain a track fastener super-pixel segmentation map; 3) constructing a track fastener superpixel undirected graph structure according to the track fastener superpixel segmentation graph; 4) respectively carrying out correlation sorting based on a background scale and a foreground scale onsuper-pixel nodes in the track fastener super-pixel graph structure; and 5) obtaining a track fastener positioning map with outstanding foreground and background inhibition according to the correlation sorting result, and completing accurate positioning of the track fastener area in the image. Compared with the prior art, the method has the advantages of accurate positioning, accurate judgment, interference inhibition and the like.

The invention discloses a multiscale infrared target adaptive detection method based on the visual reception field, relates to a multiscale infrared target adaptive detection method, and belongs to the technical field of machine vision. A calculation differential operator of an edge detector is used to calculate the gradient direction theta of an image, and an orientation information distribution map of pixel points is obtained; a local background is predicted according to background information similar to local characteristic in the infrared image; a mathematical model Gabor filter in the simple cell reception field is used to filter an object image, an object is detected preliminarily, and a direction parameter theta of the Gabor filter is determined adaptively by orientation information distribution of the pixel points; and the gray scales of the pixel points are adjusted to obtain a background inhibited and object enhanced image, and the detection precision is improved. According to the invention, the robustness, anti-interference capability and adaptability, to different scales of object, of the infrared object detection method can be improved, and the operand is low.