33results about How to "Improve detection and recognition accuracy" patented technology

The invention provides a social network site false fan detection method achieved on the basis of a network crawler by means of machine learning. The method comprises the steps that data of microblog or other social network users is automatically acquired through the web crawler, and a simulation login function is achieved; characteristic fields are selected from the data extracted by the web crawler, and a training sample and a test sample are obtained; a classical SVM algorithm classifier is adopted, multiple groups of data are extracted from the training sample randomly and guided into the classifier, and the classifier conducts machine learning to form a training classification model; the classification model is tested through the test sample, and the optimal cross validation precision is achieved by continuously adjusting setting parameters of the classification model; the microblog or other social network users are detected through the optimal classification model. Accordingly, the false fan detection precision is greatly improved, the computing amount is low, the processing speed is high, the data is not likely to be interfered in the computing process, and the method is particularly suitable for mass data processing.

The invention provides an SAR remote-sensing image oil spilling detection and identification method, which has the concrete process comprising the following steps of utilizing a Gamma MAP filter to filter an SAR image, and filtering Sobel; carrying out a watershed algorithm on a gradient map obtained through filtering Sobel to realize sea and land division; utilizing a mean value of a sea surface area image to fill a land area, and then utilizing a C-V algorithm to carry out target area division and extraction in the same homogeneous area on the filled image; extracting a gray-level co-occurrence matrix, the texture property of wavelet decomposition, the gray-level feature and the shape feature of a target area to build a vision frequency histogram; utilizing an SVM classifier model obtained through training to classify the vision frequency histogram, removing a suspected oil spilling area from the target area, and realizing initial false-alarm removal; adopting a result of false-alarm removal as an initial labeling field; utilizing a characteristic field in a context model of MRF to carry out further false-alarm removal based on the initial labeling field, so that the SAR remote-sensing image oil spilling detection and identification method is realized.

The invention discloses a cigarette brand recognition method in a complex scene. The method comprises steps of carrying out graying processing on the original color image and eliminating noise interference by combining image filtering; and utilizing an improved Sobel edge operator to roughly position the edge of the preprocessed image, obtaining a block-shaped connected candidate region of the binary image through refinement processing such as mathematical morphology operation and the like, and sending the block-shaped connected candidate region to a deep learning neural network Faster RCNN model for accurate positioning and identification. According to the method, candidate areas are intercepted through edge detection, so that the interference of the background on the detection performance is reduced, and meanwhile, an improved Sobel operator focuses on detecting the edge in the vertical direction in combination with the characteristics of shelf cigarette pictures; according to the method, the Faster RCNN detection model modifies the anchor frame size and proportion in the area suggestion network according to the cigarette size and shape characteristics, so that the missed detection probability of small targets is reduced, and the detection and recognition accuracy is improved.

The invention discloses a high-speed dimension code positioning identification system based on a full convolutional neural network. The invention discloses a high-speed dimension code positioning identification system based on a one-stage strategy full convolutional neural network. The high-speed dimension code positioning identification system comprises a data preparation module, a data enhancement module, a learning training module and a two-dimensional code detection positioning identification module. A feature extraction network of a two-dimensional code detection positioning identification module is set to be a combination of six convolution layers and five pooling layers; one pooling layer is arranged between every two convolution layers, the step length of each pooling layer is 2, the feature information of the dimension code is fully extracted to obtain a two-dimensional code feature map, and the position and the category of the input two-dimensional code are predicted in a regression mode on the output feature extraction map; and the identification system configuration of the next convolutional neural network to be tested is automatically adjusted and reconstructed according to the test effect of the network model, so that the real-time performance is enhanced. According to the positioning recognition system, the types and the position coordinates of one or more two-dimensional codes in the picture can be detected at the same time, the detection recognition precision is higher than 95%, and the detection speed is lower than 5 ms/frame.

The invention relates to a power transmission corridor foreign matter detection method and system based on a twin network, and belongs to the technical field of remote sensing. The method comprises the following steps: firstly, acquiring a multi-temporal remote sensing satellite image of a power transmission corridor, carrying out artificial fine labeling on a change target and a region in the multi-temporal image, then slicing and binarizing the labeled region, then training a training set containing foreign matters in the power transmission corridor by using a twin network, obtaining a modelcapable of detecting the foreign matters in the power transmission corridor, and then detecting the multi-temporal remote sensing satellite image of whether the foreign matters exist in the power transmission corridor or not by using the obtained model to judge whether the foreign matters exist in the power transmission corridor or not. The method can detect and identify various power transmission corridor foreign matters, has calculation efficiency significantly superior to that of a manual method, has higher foreign matter detection accuracy and reliability, and provides convenience for operation and maintenance of a power transmission line.

The invention provides an action recognition method and device and electronic equipment, and relates to the technical field of image processing, and the method comprises the steps: obtaining a plurality of images containing a target object and optical flow images of the plurality of images if the target object is detected from a video frame; extracting an object trajectory feature of the target object from the plurality of images, and extracting an optical flow trajectory feature of the target object from the optical flow images of the plurality of images; and identifying the action type of the target object according to the object trajectory feature and the optical flow trajectory feature. According to the embodiment of the invention, the action type of the target object is identified according to the track information of the target object in the video frame and the optical flow information of the target object in the optical flow image of the image, and the time feature information and the space feature information of the target object are fused, so that the detection and identification precision of the action type is effectively improved, the detection efficiency can be considered, and the overall detection performance is improved.

The invention relates to an automatic change detection method for a remote sensing image in a large-scale complex scene. The method comprises the following steps: S1, acquiring a remote sensing imagedata pair of front and back time phases; S2, extracting feature points from the remote sensing image data pair, and performing image registration; S3, obtaining a difference image through a differencemethod based on the registered remote sensing image data pair; S4, extracting the saliency of the difference image, and generating a variable pseudo training sample and an invariable pseudo trainingsample; and S5, inputting the variable pseudo training sample and the invariable pseudo training sample into a classifier, performing binary classification on the difference image obtained in step S3,and obtaining binary detection results about the variable type and the invariable type. Compared with the prior art, the method can be applied to change detection of remote sensing images in large-scale complex scenes, and has the advantages of high detection and recognition precision, high efficiency and the like.

The invention relates to an intelligent signal lamp control method based on urban major and collector road intersections. Major road entrance lanes of the intersections can allow turning left, going straight and turning right and are provided with earth induction coils A1, sensors B01, sensors B03 and sensors C, major road exit lanes of the intersections are provided with earth induction coils A2,data acquired by the earth induction coils A1, the earth induction coils A2, the sensors B01, the sensors B03 and the sensors C are processed and analyzed a microprocessor M, and prolonged time of green lamps of major roads of the intersections is determined according to a delay method of the green lamps. According to the method, timing is achieved by sufficiently utilizing signals, vehicles of the intersections are evacuated, time is saved, and jam is decreased.

The invention discloses a dual spectrum technology based leaf and stalk separating and cutting quality detection device and detection method. The detection device comprises a spreading device (1), a paving machine (2) arranged under the spreading device (1), an X ray detector (3) arranged under the paving machine (2), a visible light detector (4) arranged under the outlet of the X ray detector (3), a receiving device (6), a frame (5), and a control cabinet (7). The receiving device (6), the frame (5), and the control cabinet (7) are arranged under the outlet of the visible light detector (4).By adopting a visible light and X ray imaging technology and corresponding image algorithm, the tobacco leaf area, tobacco stalk area, and tobacco stalk length can be obtained; the leaves and stalks in a same image do not need to be separated, and the identification accuracy of image detection is improved. The leaf structure and long stalk rate can be detected, the stalk containing rate of leavesand leaf containing rate of stalks can be detected, the stalk density and leaf density are not needed during the index calculation process, and the leaf cutting quality indexes can be automatically calculated out.

The invention provides a method for facilitating bridge automatic detection and identification, and a bridge detection and identification tag, and belongs to the technical field of traffic engineering. The method comprises the following steps of (1) collecting information of bridge members and information of surrounding environment objects; (2) selecting a tag used for a bridge; (3) storing the information of the bridge members, the information of the surrounding environment objects and positions of surrounding tags in the tag; and (4) arranging the tag on the bridge, wherein the tag has a communication interface relative to a detection apparatus, thereby allowing the detection apparatus to obtain the information carried by the tag. The method is easy to realize, relatively low in resource consumption, high in identification accuracy and high in navigation and locating precision.

The invention discloses a rice tillering stage weed segmentation identification method based on an improved coding and decoding network. The method comprises the following steps: (1) collecting a rice farmland image in a tillering stage; (2) carrying out image preprocessing, image enhancement and semantic annotation on the collected rice farmland image; (3) inputting the image subjected to preprocessing enhancement and semantic annotation into the improved coding and decoding U-Net network, and training the improved coding and decoding U-Net network to obtain a rice weed segmentation model; and (4) performing identification detection on a to-be-detected rice farmland image in a tillering stage by using the rice weed segmentation model, and outputting an identification detection result to obtain segmentation conditions of rice and weeds in the rice farmland image. According to the invention, accurate identification and accurate positioning of the weed-dense area in the paddy field are realized by adopting the improved coding and decoding network, so that the accurate herbicide spraying operation in the weed-dense area in the paddy field can be guided, and the pesticide dosage is reduced.

The invention relates to a video image distortion effect model construction method based on an improved dice loss function. The method comprises: firstly, performing function improvement based on a Dice loss function, and a weight factor and a smoothing factor are added to better adapt to the characteristics of a sample data set; secondly, training data in the dense convolutional neural network ofthe DenseNet by adopting an improved loss function to realize classified construction of the model; and finally, carrying out classification prediction on the existing video image by using the trained model, and judging whether the video image is distorted or not. Compared with a traditional loss function training model, the identification and detection precision of the improved loss function insix common video image distortion effect data sets is improved, and the advantages are obvious.

The invention relates to the technical field of computer vision, and particularly provides a target vehicle detection method and device. The target vehicle detection method comprises the steps that a to-be-processed image is acquired, and the to-be-processed image comprises a vehicle; performing image detection on the to-be-processed image to obtain a detection result of the target object; the target object is at least one type of vehicle component belonging to the vehicle; and determining that the vehicle in the to-be-processed image is a target vehicle in response to the condition that the detection result of the target object meets a preset condition. According to the embodiment of the invention, the false detection and missing detection risks of the target vehicle are reduced, and the detection and recognition precision is improved.

The invention provides a remote sensing image processing method and device and a server. The method comprises the steps: after a target remote sensing image for a target area is obtained, processing the target remote sensing image firstly to obtain a corresponding target gray level image; carrying out histogram-based equalization processing on the target grayscale image according to a preset equalization processing rule to obtain an equalized target grayscale image; and detecting a target building in the target area according to the equalized target grayscale image. The histogram-based equalization processing is performed on the target grayscale image according to the preset equalization processing rule, so that the equalized target grayscale image with a large grayscale dynamic range, a high contrast ratio and richer image details can be obtained; and then, the target building in the target area can be accurately detected and identified from the remote sensing image based on the equalized target grayscale image, so that the detection error is reduced, and the detection precision is improved.

The invention relates to the field of computer vision, adversarial learning and deep learning target detection, in particular to a multi-class plankton detection method under the condition of unbalanced plankton population distribution. According to the invention, on one hand, a cyclic adversarial neural network structure is integrated, and image sample data of a non-dominant distribution population is expanded, therefore, the sample data of various groups of images are balanced in quantity, the over-fitting phenomenon of the detection model in training is avoided, and the overall recognition precision is improved. And on the other hand, the algorithm introduces a DenseNet structure with a feature reuse characteristic on the basis of the YOLOV3 model to replace a down-sampling layer in the original YOLOV3 model, so that the feature loss of the fine features of the plankton in the transmission process of the deep neural network layer is reduced, and the stability of feature propagation is improved.

The invention discloses a video-based live ammunition and laser dual-mode target scoring system and a target scoring method. The target scoring system is of a box type structure and comprises an image acquisition module, an image processing module, a wireless communication module and a power supply module before a target plate is arranged; the image acquisition module is connected with the image processing module, the image processing module is connected with a target scoring terminal through the wireless communication module, and the power supply port of the power supply module is respectively connected with the power supply ends of the image acquisition module, the image processing module and the wireless communication module. The target scoring method sequentially comprises the following steps of image acquisition, image processing and transmission of an image processing result to the target scoring terminal. The system has the characteristics of portability, simple deployment, high precision and the like. According to the method, the problems of danger and low efficiency of manual target scoring are solved, and the military training efficiency of troops is improved. The system is suitable for any live ammunition or laser shooting training field.

The invention aims to provide a foreign matter detection method for an LC series topology wireless charging system. The method is relatively low in hardware cost, high in reliability and relatively high in precision. The method comprises the following steps that: step 1, in a primary power-on calibration stage, after power-on, the natural resonant frequency f, the quality factor Q and the equivalent direct current resistance of an LC network of a transmitting end are calculated, and recorded as common parameters; and 2, the output current If of a coupled current sensor is collected in real time, and the effective value Uf of the input voltage of an excitation source and the current working frequency f1 of the excitation source during sampling are recorded, the parameters are substituted into a quality factor change rate conversion formula, so that the change rate of the quality factor Q caused by the intervention of metal foreign matters is to calculated,if the change rate of the current quality factor Q is smaller than a set metal foreign matter detection judgment threshold value, an alarm is given, and charging is stopped, and if the change rate of the current quality factor Q is within the set foreign matter detection judgment threshold range, a charging system is safe, and charging is continued.

The invention discloses an intelligent article searching method for visually impaired people. The method comprises the following steps: 1, extracting keywords of article appeals required by the visually impaired people through a voice recognition model based on deep learning; 2, constructing a target detection model based on deep learning; 3, collecting a life article data set for visually impaired person target detection and training a target detection model; and 4, after searching the target, outputting the position information of the article appealed by the visually impaired person through voice. According to the invention, the voice appeal of the visually impaired person can be identified through artificial intelligence, and the article can be identified and positioned intelligently, so that the visually impaired person can be helped to search the required article.