39results about How to "Guaranteed retrieval efficiency" patented technology

The invention discloses an image retrieval method based on variable-length depth hash learning and mainly relates to the field of image retrieval and depth learning. According to the method, learning of hash codes is modeled into the process of similarity learning. Specifically, the method utilizes a training image for generating a batch of ternary image sets. Each ternary image set comprises two images with the same label and one image with the different label. The purpose of model training is to space image pairs matched to the maximum and the unmatched image pairs in hamming space. The depth convolution nerve network is introduced into the learning part of the method, the image characteristics and hash functions are optimized in a combined mode, and the end-to-end training process is achieved; hash codes output by the convolution network have different weights. For the different retrieval task, a user can regular the length of the hash codes by disconnecting unimportant bits; meanwhile, according to the method, the discrimination of the hash codes can be kept effectively under the circumstance that the hash codes are short.

The invention relates to a trademark image retrieval method which includes: performing color and local feature extraction on sample trademark images; categorizing the sample trademark images according to grey level frequency; performing foreground and background segmentation; determining the edge criteria according to foreground and background pixel gray level statistical values; extracting edge features according to the edge criteria; judging the complexity of the sample trademark images according to the edge features so as to divide the sample trademark images into at least one group; using the same steps to extract the color features, edge features and local features of a to-be-retrieved trademark image; judging the complexity of the to-be-retrieved trademark image according to the edge features of the to-be-retrieved trademark image to determine the group which the to-be-retrieved trademark image belongs to; performing similarity retrieval in the group which the to-be-retrieved trademark image belongs to by synthesizing the color and edge features of the to-be-retrieved trademark image; performing similarity retrieval in other groups when the fact that the to-be-retrieved image is not similar to each sample trademark image in the group which the to-be-retrieved trademark image belongs to is determined; performing similarity retrieval according to the local features of the to-be-retrieved trademark image when retrieval needs to be performed again.

The invention relates to the technical field of artificial intelligence, in particular to an image retrieval method and device, equipment and a storage medium. The image retrieval method comprises the steps of obtaining a to-be-recognized image; inputting the to-be-recognized image into a pre-trained image clustering model, and predicting a target class cluster identifier corresponding to the to-be-recognized image, wherein the image clustering model divides a pre-created image library into a plurality of pre-divided class clusters; obtaining a pre-divided class cluster corresponding to the target class cluster identifier from the plurality of pre-divided class clusters as a target retrieval class cluster, wherein the target retrieval cluster comprises at least one to-be-matched image; and performing image retrieval on the to-be-recognized image based on the target retrieval cluster. According to the method, the clustering precision can be effectively ensured, and the problems of large calculation amount and time consumption caused by traditional image retrieval can be effectively avoided.

The invention provides indexing equipment which comprises a numeric string segmentation module and an indexing establishment module, wherein the numeric string segmentation module is used for segmenting a numeric string in a document and the indexing establishment module is used for establishing an index based on a new numeric string obtained after the numeric string is segmented and the document. Due to the fact that the technical scheme includes that the numeric string in the document is proposed to be segmented and the new segmented numeric string is used as a keyword to establish the index, the indexing equipment has the advantages of being capable of reasonably controlling the length of a file list pointed by the keyword in an index structure and the length of a keyword list, namely, controlling the size of the indexing established by the numeric string within the acceptable range of computer memory and improving indexing speed of the numeric string. The invention further provides an indexing method, a search device, a search method and a search system.

The invention discloses a rapid audio frequency comparison method. The audio frequency comparison part of the rapid audio frequency comparison method comprises the following steps of: reading an audio frequency p and an audio frequency q and marking off a feature segment set Cp of the audio frequency p and a feature segment set Cq of the audio frequency q; rapidly calculating an energy feature value sequence Wp of the feature segment set Cp of the audio frequency p and an energy feature value sequence Wq of the feature segment set Cq of the audio frequency q by utilizing a CUDA (Compute Unified Device Architecture) function on a GPU (Graphics Processing Unit); forming an energy matrix by a feature value of each feature segment according to the sequence of the feature segments; finding out a common feature segment set Seg of two feature value sequences by utilizing a suffix array deformation algorithm; rapidly scanning the common feature segment set Seg, finding out a communication region and using a set Vres returning to the communication region as an audio frequency comparison result; and marking the comparison result on an oscillogram.

The invention relates to a footprint image retrieval method based on a multi-scale local attention enhancement network. Compared with the prior art, the defects that footprint image retrieval is low in accuracy and slow in speed are overcome. The method comprises the following steps: obtaining and preprocessing a training sample; constructing a footprint image retrieval model; training a footprint image retrieval model; obtaining a footprint image to be retrieved; and obtaining a footprint image retrieval result. According to the method, more effective fine-grained features can be extracted, the feature representation capability of the image is further mined, and the footprint image retrieval speed and accuracy are improved.

The invention discloses a crawler system IO optimization method and device, relates to the field of software engineering, and aims at solving the problem that existing result storage work carried outby taking crawler tasks as units is low in IO efficiency and influences the retrieval efficiency. The method comprises the following steps of: caching received first crawlers by a first result processor, and when the fact that the quantity of cached crawling results exceeds an aggregation threshold value is determined, writing the plurality of crawling results into an aggregation file according toan end-to-end splicing method and recording a position offset of each crawling result; generating an aggregation path stored in a big file object storage system according to a content of the aggregation file and sending the aggregation file to the aggregation path; and generating an aggregation log which comprises each crawling result, the position offset of each crawling result, the aggregationpath and a number of each crawler according to the aggregation file and sending the aggregation log to a log processor.

The present invention discloses a keyword searchable encryption method based on a Bloom filter with a storage structure, which the following steps: step 1, system parameter initialization; step 2, user secret key generation; step 3, retrieval index generation; and step 4, search token generation; step5,cloud storage server search on cryptograph key word. The present invention allows users to use trap door search linking with a key word to encrypt a document, allows data users to encrypt their data and store the data in the cloud storage server, when needed, the needed cryptograph data can be searched by a key word search token, and then is loaded and decoded. The present invention solves the problem of low retrieval efficiency of a conventional encryption method in the prior art.

The invention discloses a subtitle matching method and system based on image retrieval. By acquiring the image to be matched, the sample image and the subtitle content corresponding to the sample image, the collected sample image is used as the data for ELM self-encoding network training, and the subtitle content to be matched is The image and each sample image are respectively trained with the ELM self-encoding network to obtain a hash function, and the hash function is used to convert the binary code of the image to obtain the binary code sequence of the image to be matched and each sample image, and calculate the Chinese between the calculated images. According to the Hamming distance, similar image retrieval is carried out by the K nearest neighbor method to obtain the most similar sample image with the image to be matched, and finally the subtitle content corresponding to the most similar sample image is given to the image to be matched; The accuracy of image similarity retrieval, and the use of binary coding ensures the efficiency of retrieval and achieves the effect of fast and accurate image subtitle matching.

The invention discloses an image searching method, which comprises a training part and a searching part, wherein the training part comprises the following steps of: the extraction of characteristic points, the supplementation of the characteristic points and the determination of matching relationships, the generation of similar point set, the clustering of the characteristic point sets and the generation of characteristic vectors of each image in an image database; and the searching part comprises the following steps of: extracting the characteristic points of a picture to be retrieved and generating the characteristic point sets; calculating distances between each characteristic point descriptor vector and corresponding cluster centers, and determining a cluster where a current characteristic point belongs by using a smallest distance; calculating the frequency ni of each cluster where the characteristic points of the picture to be retrieved belong; based on the frequency ni of the clusters where the characteristic points of the picture to be retrieved belong, and the probability logarithm wi of each cluster, generating and unitizing the characteristic vector; and calculating Euler distances between the characteristic vector of the picture to be retrieved and the characteristic vectors of each image in a picture library, and selecting the image output with the smallest distance as a searching result.

The embodiment of the invention discloses a retrieval method and a retrieval device. The method in the embodiment of the present invention includes: obtaining a retrieval request, the retrieval request including retrieval characters; judging whether the retrieval request is to perform intelligent association on the retrieval characters or to perform content retrieval; when it is determined to perform intelligent association on the retrieval characters, Retrieving the search character in a specific field of the index database; when it is determined that the content search is performed on the search character, retrieving the search character in the index database.

The invention provides an implementation method for a mass data full-text retrieval scene and belongs to the field of mass data full-text retrieval. According to the method, Lucene is introduced into a retrieval engine, and Lucene index files are established for data files existing in a cluster; a daemon is set on each node storing the corresponding Lucene index file to maintain the Lucene index file on the node. Through the method, a scheduling mechanism of fragments by a coordinator is optimized, counting judgment is performed on each execution node, and the execution location of each fragment is adjusted to balance node resources; a full-text reading engine mechanism is also optimized, Lucene is preferentially adopted to perform retrieval, and RCFile retrieval is called in case of a problem to guarantee the continuity and correctness of retrieval; retrieval performance is improved, and cluster resources can be utilized more reasonably.

The invention discloses a high-dimensional space hypersphere covering method for human motion capture data retrieval, which relates to the technical field of human motion capture, in particular to a method for expressing, extracting and retrieving characteristics in optical human motion capture data. The method introduces a distribution characteristic concept and a method of various motions of a human body in a high-dimensional space and comprises the following steps: extracting the main characteristics in human motion data; constructing characteristic vectors; by aiming at the characteristicvector of a retrieving condition (namely target motion to be retrieved), constructing a series of hypersphere covering area in the corresponding high-dimensional space; defining covering sequences and then evaluating the similarity of the covering sequences; and realizing the retrieval of the motion capture data. The method can effectively realize human motion capture data retrieval by a practical test and has favorable practicability and robustness.

The invention discloses a rapid audio frequency comparison method. The audio frequency comparison part of the rapid audio frequency comparison method comprises the following steps of: reading an audio frequency p and an audio frequency q and marking off a feature segment set Cp of the audio frequency p and a feature segment set Cq of the audio frequency q; rapidly calculating an energy feature value sequence Wp of the feature segment set Cp of the audio frequency p and an energy feature value sequence Wq of the feature segment set Cq of the audio frequency q by utilizing a CUDA (Compute Unified Device Architecture) function on a GPU (Graphics Processing Unit); forming an energy matrix by a feature value of each feature segment according to the sequence of the feature segments; finding out a common feature segment set Seg of two feature value sequences by utilizing a suffix array deformation algorithm; rapidly scanning the common feature segment set Seg, finding out a communication region and using a set Vres returning to the communication region as an audio frequency comparison result; and marking the comparison result on an oscillogram.

The invention discloses an AOI visual inspection system and a detection method thereof. The system comprises a machine body and a control box, a moving mechanism is arranged at the lower end of the machine body, a display module is fixedly arranged at the front end of the machine body in an inlaid mode, an input module is fixedly installed at the front end of the machine body, and a protection mechanism is arranged at the front end of the machine body; a permanent magnet is fixedly installed at the front end of the machine body, a shielding mechanism is arranged at the front end of the machine body, a detection mechanism is arranged in the machine body, an inlet is formed in the front end of the machine body, and a placement plate is movably installed in the inlet. During detection, the shielding plate slides downwards along the sliding groove, so that the shielding plate covers and seals the inlet, interference of external light to a detection result is prevented, the first camera moves into the placement groove, the shielding plate is moved upwards to reset after detection is completed, and the shielding plate can be attracted and fixed to the lower end of the permanent magnet through magnetic force.

The invention discloses a crawler network path tracing method based on object storage. The method comprises the steps that an object storage system and a log recorder are established, wherein the log recorder generates a result path log, and indexes from the source URL of a crawling result to a crawler result file on the object storage system are recorded in the result path log; when an external system needs to call the data in the database, the crawler result file on the object storage system is directly obtained through the indexes. According to the method, the object storage system is introduced so that the file reading and writing speed can be increased; the result path log is established so that the data can be retrieved in the log when the external system calls the data and does not need to be searched in the database, and accordingly the possibility of reading and writing conflicts is avoided.