245results about How to "Improve detection results" patented technology

The invention discloses a traffic identifier detection method based on multi-scale circulation attention network. The method comprises the following steps: firstly, building a traffic identifier detection model, wherein the traffic identifier detection model is formed by compounding a convolutional neural network model feature extraction model for carrying out image feature extraction and a multi-scale circulation attention network model for improving small-target detection accuracy; then training the traffic identifier detection model by utilizing a reasonable training sample so as to acquirea trained traffic identifier detection model; and inputting to-be-detected images into the trained traffic identifier detection model during testing so as to acquire a detection result. According tothe method disclosed by the invention, by applying an encoder/decoder structure, the acquired features are enhanced, small targets are detected by using a multi-scale attention structure, and referring to a residual difference structure, the problems of gradient disappearance and gradient explosion are solved. Compared with the other advanced traffic identifier detection methods, the method disclosed by the invention has the advantage of competitiveness.

The invention discloses a target space knowledge and two-stage prediction learning-based target detection method. By utilizing various data conversion methods, a sample number is increased and samplediversity is improved; two deep neural networks including SSD and newly designed RefineNet are trained; for a prediction target with a relatively high probability in a primary prediction result of theSSD, the accuracy of judgment is further improved through the RefineNet; and by establishing peculiar spatial structure constraint rules of the target, the wrong prediction is reduced, thereby obtaining a final detection result. Compared with a few existing methods, the method provided by the invention has the advantages that visual and spatial characteristics of a remote sensing target are considered at the same time; and end-to-end target candidate selection, feature extraction and classified locating are realized by utilizing the deep networks with excellent feature extraction capabilities, so that the detection rate of the remote sensing target is remarkably increased and the false alarm rate is reduced.

The invention discloses a device used for quick online nondestructive detection on cracks of poultry eggs on the basis of acoustic characteristic and a method thereof; the poultry egg is arranged on a supporting idler wheel; a stepping motor is connected with the supporting idler wheel and drives the supporting idler wheel; a microphone which is arranged beside the poultry egg is sequentially connected with a computer by a signal amplifier, a filter and an A/D converter; when the poultry egg rolls on a chain, different parts of the poultry egg are mechanically knocked at multi-point by a knocking rod; sound signals collected by the microphone are amplified by the signal amplifier, filtrated by the filter and converted by the A/D converter and subsequently input to the computer, thus comprehensively judging the cracking situation of the poultry eggs. The device and the method can adapt to poultry eggs with different types and different sizes, realize the classification and automatic detection of the poultry eggs during the production processing and circulation process, and can be popularized and applied to the nondestructive detection on the hardness and the maturity of agricultural products such as fruits, vegetables, and the like.

The invention discloses a fast pedestrian detection method and device. The method comprises the following steps: step S1, constructing a configurable deep model based on a convolutional neural network, and utilizing training samples to obtain parameters of a constructed network by learning to obtain a model used for a test process; and step S2, inputting test samples, utilizing a variation law ofneural-network perception domains, using different intermediate layers to detect target objects in different scale ranges through the trained model, and obtaining box graphs of the target objects in images by prediction. The method uses the different intermediate layers to detect the target objects in the certain scale ranges through utilizing the variation law of the neural-network perception domains, better adapts to relationships of the perception domains and object sizes, and effectively improves a detection result.

The invention provides a method for positioning PU of cognitive radio networks. The method comprises the following steps: acquiring a plurality of detecting results on the PU based on MAC communication protocol through an idle channel capable of perceiving at least three SU of the PU within certain time slot; and based on the plurality of the detecting results in the last step, calculating and determining the position of the PU by an iterative algorithm of weighting least squares. The method has the advantage that the method acquires location algorithm-based positioning precision by utilizing a little information without basing on the location algorithm, and further improves the performance of positioning algorithm by introducing the iterative algorithm of weighting least squares. The deduction of Cramer-Rao theory performance limit and a plurality of imitations prove that the positioning error of the method is extremely similar to Cramer-Rao limit theory performance.

The invention relates to an automatic driving scene key target detection and extraction method based on deep learning, and is characterized in that the method carries out key target detection, semantic annotation and target frame selection in a driving scene based on an improved single multi-frame detector, and the training process of the improved single multi-frame detector specifically comprisesthe steps: 1) obtaining a training set; 2) generating a prior frame, and matching the prior frame with a real target with class labels and frame labels of corresponding pictures to form positive andnegative samples; And 3) iteratively updating the model parameters of the single multi-frame detector by utilizing the loss function, wherein the iterative updating is realized based on momentum optimization gradient updating. Compared with the prior art, the method has the advantages of high detection precision, capability of detecting a small target and the like.

The invention belongs to the technical field of computer vision, and particularly relates to a multi-target tracking method. The shielding and appearance similarity problems in multi-target tracking still limit the performance of a visual multi-target tracking algorithm. The invention provides a multi-target tracking method. The method includes: detecting each frame of image in the video; outputting a detection result and constructing a nonlinear motion model based on an LSTM network; and constructing a multi-target tracker, outputting a prediction result of the multi-target tracker, constructing a data association module based on a Hungary algorithm, inputting a tracking target prediction position and a feature vector matrix, outputting an allocation probability vector to obtain a targetdetection result with the maximum matching probability, and taking the target detection result as a tracking result of the ith target. The problems of inaccurate tracking and identity switching aftershielding in an existing visual multi-target tracking algorithm are solved, and the tracking performance is greatly improved.

The invention discloses a method for automatically judging a dress code through a surveillance video, and is mainly used for judging whether the dressing of the head of an operator is normative or not. The method comprises the following steps that: through a camera, collecting the head characteristics of the operator on line, and generating an image; and through a deep learning training model, extracting image characteristics, judging whether the dressing of the head of the operator is normative or not, and if the dressing of the head of the operator is not normative, outputting alarm information. By use of a deep learning method, features are automatically extracted, the extracted features exhibit better robustness so as to greatly improve a detection result, working efficiency is improved, supervision cost is saved, a supervision dimension is expanded, and supervision accuracy is improved.

The invention belongs to the technical field of image target detection, and particularly relates to a Faster R-CNN-based bridge crack instance segmentation method, which comprises the steps of 1, constructing a bridge crack data set; 2, marking a training sample; 3, building a bridge crack instance segmentation model of the improved Faster R-CNN; 4, training the instance segmentation model built in the step 3; 5, testing the instance segmentation model trained in the step 4; 6, performing actual detection. Compared with the prior art, the invention is higher in robustness, accurate bridge crack classification and positioning results can be obtained, and a high-quality bridge crack segmentation mask can be generated and used for evaluating the damage degree of a bridge and making a corresponding maintenance scheme; in addition, the method can achieve the accurate detection of a plurality of cracks in the image, and can improve the detection efficiency and obtain a complete crack form incombination with the image splicing technology.

The invention discloses a Sobel edge detection and image block brightness feature-based blind image tampering forensic method. The method is characterized by comprising the following steps of: converting a to-be-detected suspicious image into a grayscale image; carrying out convolution processing on the grayscale image I (i, j); obtaining a gradient image G (i, j) of the image; carrying out threshold value segmentation on the G (i, j) to obtain the gradient image G (i, j); carrying out binary processing on the gradient image G (i, j) to obtain a binary image W (i, j); carrying 1 pixelation on the binary image; and judging the similarity of two sub-image sets. According to the method, blocking processing is carried out on image sets; through brightness mean value sorting, the forensic algorithm efficiency can be effectively improved; and through comparing the similar brightness values of image blocks, the correctness is further improved and the image forensic efficiency is further improved. According to the method, the problem that the tampered images cannot be correctly detected due to cloning and tampering behaviors of large-scale zooming is solved; and through normalized image block brightness values, the detection result of cloned images with different brightness values is further improved.

The invention relates to an image processing method and device, electronic equipment and a storage medium. The method comprises the steps: obtaining an iris image group which comprises at least two iris images to be compared; detecting an iris position in the iris image and a segmentation result of an iris region in the iris image; performing multi-scale feature extraction and multi-scale featurefusion processing on an image region corresponding to the iris position to obtain an iris feature map corresponding to the iris image; and executing comparison processing by using the segmentation results and the iris feature maps corresponding to the at least two iris images, and determining whether the at least two iris images correspond to the same object based on a comparison result of the comparison processing. According to the embodiment of the invention, accurate comparison of iris images can be realized.

The invention discloses an image saliency detection method and device based on correlation constraint graph sorting. The method comprises the steps of performing super-pixel segmentation on a to-be-detected image, building a closed-loop graph model, and then calculating prior information of each super-pixel node; extracting information of color, texture, position and the like of an input image; obtaining a foreground probability value of each super-pixel node; taking a set of the nodes with the foreground probability values greater than a first preset threshold value as a foreground seed pointset ind_fore; taking a set of the nodes with the foreground probability values smaller than a second preset threshold value as a background seed point set ind_back, wherein the first preset thresholdvalue is greater than the second preset threshold value; and performing calculation by using a correlation constraint graph sorting model to obtain a foreground probability S_f of each super-pixel node, and using a foreground probability value S_f as a final saliency estimation value S_final. By applying the image saliency detection method and device, a saliency detection result can be more accurate.

The invention provides a kit. The kit comprises a probe, wherein the probe is fixed to a solid-phase substrate or floats in a solution, and can specifically recognize a gene area of at least ten genes among 60 genes in the table 1. The invention further provides applications of the kit, a method for constructing a target area sequencing library, a sequencing method, a method for detecting the target area variation, a device for detecting the target area variation, as well as a method and a device for carrying out early screening on the colorectal cancer. With the adoption of the kit and/or the methods and the devices provided by the invention, the related gene sequences of the colorectal cancer can be easily and conveniently obtained in a high-specific manner at one time, and accurately detected and analyzed, and a detection and analysis result can be used for assisting the early screening judgment or monitoring on the colorectal cancer.

The invention discloses a video area removal tampering detection method and device. According to the method, the GIN network model of the seven-layer structure is constructed, the GIN network model istrained and optimized, the video data set to be detected is input into the optimized GIN network model to be detected, the detection result is high in accuracy and applicability, and the technical problems that an existing video tamper detection method is not high in detection accuracy and low in applicability are solved.

The invention discloses a multi-scale deep convolutional neural network-based satellite image target detection method. The method comprises the steps of: collecting a satellite image training data set, and carrying out sample labeling; preprocessing the satellite image training data set; building a multi-scale deep convolutional neural network; inputting the preprocessed training data set into a target detection framework based on the multi-scale deep convolutional neural network to perform training so as to obtain a trained target detection neural network; and inputting a to-be-detected satellite image set, performing target detection by adopting the trained target detection neural network, and outputting an identification result. With the method adopted, the accuracy of the result of thedetection of fine-grained characteristics by the network is improved; the capability of the network in distinguishing different objects is improved; and the effect of the detection of small objects and dense object groups is improved. The method has higher robustness and can effectively improve target detection efficiency and lower hardware requirements.

The invention discloses a chitosan quaternary ammonium salt couplant and a preparation method thereof. The chitosan quaternary ammonium salt couplant is prepared from 0.1-0.9wt% of Carbomer 941, 0.5-6wt% of phenoxyethanol, 0.01-1.0wt% of hydrochloric acid, 0.1-0.8wt% of chitosan quaternary ammonium salt, 1.2-10wt% of triethanolamine, 2-20wt% of a humectant and the balance distilled water. The chitosan quaternary ammonium salt couplant has the advantages of good bactericidal effects, good environmental performances, no carcinogenesis on the human body, clarification, transparency, alcohol fragrance, good homogeneity and good fineness.

The invention belongs to the technical field of image target detection, and particularly relates to a bridge crack detection method based on a multi-resolution convolutional network, which comprises the following steps: 1, constructing a bridge crack data set; 2, labeling and classifying the training samples; 3, building an improved HRBCS bridge crack semantic segmentation model based on the HRNetmodel; 4, training the semantic segmentation model built in the step 3; 5, testing the semantic segmentation model trained in the step 4; 6, performing actual detection. Compared with the prior art,the method has the advantages that the spatial accuracy is greatly improved, a high-resolution segmentation result can be obtained, segmentation of small cracks and complex cracks is greatly improved,different types of cracks can be detected and classified, and a high-quality bridge crack segmentation mask can be generated; therefore, a corresponding maintenance scheme is formulated according tothe damage degree of different types of crack evaluation to the bridge.

The invention provides spindle sleeve part inner/outer diameter detection equipment. The spindle sleeve part inner/outer diameter detection equipment comprises a work station, a pressure spring, a positioning sleeve, a sliding sleeve, a positioning mandrel, an inner diameter sensor probe, an outer diameter sensor probe and a handle, wherein the positioning sleeve and the positioning mandrel are mounted on the work station, the sliding sleeve is arranged between the positioning sleeve and the positioning mandrel, clearance fit is realized externally and internally, the positioning mandrel is arranged at a center of the sliding sleeve, a lower end of the sliding sleeve is provided with the pressure spring, the outer portion of the sliding sleeve is the positioning sleeve, the positioning sleeve is provided with a limit ring groove in a vertical direction, the limit ring groove is internally provided with the handle, the handle is connected with the sliding sleeve, and the inner sensor probe and the outer sensor probe are arranged on the positioning mandrel and the positioning sleeve. The detection equipment has properties of high display value precision and good detection data consistency, an inner diameter, an outer diameter, inner/outer circle roundness and conicity of workpieces can be detected through one-times operation, manpower cost is low, and the detection equipment is suitable for detecting dimensions, roundness and conicity of spindle sleeve circular ring products which are made through batch-scale production and have high inner-hole and outer-circle precision requirements.

The invention provides a kit, which comprises a probe, wherein the probe is immobilized on solid-phase substrate or the probe is free in a solution; and the probe is capable of achieving specific recognition on gene regions of at least 40 genes in 145 genes in List 1. The invention also provides application of the kit, a method for constructing a target region sequencing library, a sequencing method, a method for detecting target region variation, a device for detecting the target region variation as well as a method and a device for early screening of lung cancer. By virtue of the kit and/or the method or device provided by the invention, related gene sequences of the lung cancer can be simply and conveniently acquired with high specificity in one time, and the related gene sequences can be accurately detected and analyzed, so that a detection analysis result can be used for assisting in judging or monitoring the early screening of the lung cancer.

The present invention provides an automatic heart disease detection system based on a convolutional neural network. The automatic heart disease detection system comprises a database, a lesion information processing module, a key frame extraction module, a data cleaning module, an information fusion and storage module, a convolutional neural network learning module, a model storage and screening module and a lesion detection module. The invention relates to the fields of deep learning, medical treatment and computer vision. By design of scientific optimization, a set of automatic intelligent detection system for coronary artery calcification, complete occlusion and thrombosis based on the convolutional neural network is created and achieved, aiming at the requirements of diagnosis on resultaccuracy and timeliness, the design structure is ingenious, a high-quality auxiliary diagnosis detection result can be output without any manual assistance, and convenience is brought to coronary artery patients and doctors.

The invention discloses a block diagonalization and low-rank representation-based hyperspectral camouflaged target detection method, which is used for solving the technical problem of low target detection efficiency of an existing hyperspectral camouflaged target detection method. According to the technical scheme, the method comprises the steps of firstly dividing a background into different types by utilizing a k-means clustering algorithm; secondly according to a clustering result, sorting original data; thirdly obtaining a dictionary of each type by utilizing a PCA dictionary learning algorithm, and then obtaining a global background dictionary; fourthly according to a low-rank and sparse representation theory, building a block diagonalization and low-rank detection model; fifthly after model solving, dividing the original data into a background part and a sparse part containing a camouflaged target; and finally extracting the camouflaged target from the sparse part. Under a framework of the low-rank and sparse representation theory, the background is subjected to refined description by utilizing the clustering algorithm, so that the background description is more accurate andthe camouflaged target detection efficiency is improved.

One or more embodiments of the invention provide an optical cable fault positioning method and an optical cable fault positioning device based on deep learning, and equipment. The optical cable faultpositioning method comprises the steps of: acquiring a to-be-detected optical fiber of an optical cable; sequentially acquiring a plurality of pieces of Brillouin frequency shift information of the to-be-detected optical fiber according to the wiring direction of the to-be-detected optical fiber by using BOTDR equipment; sequentially inputting the plurality of pieces of Brillouin frequency shift information into a pre-trained fault detection model to obtain a plurality of identification results corresponding to the plurality of pieces of Brillouin frequency shift information; if the identification result is that the to-be-detected optical fiber is at a fusing position, positioning the position of a tower; and positioning the to-be-detected optical fiber of which the identification result is that the to-be-detected optical fiber is at the fusing position according to the tower position. According to the optical cable fault positioning method and the optical cable fault positioning device, the light operation state information acquired by the BOTDR equipment can be automatically identified, the labor cost is reduced, the optical cable fault can be accurately positioned, and the fault detection model can identify various fault types and position the faults.