61 results about "Artificial intelligence" patented technology

The embodiment of the invention discloses a shooting method. The shooting method includes the steps that when a request message for focusing a target object is received, focusing is performed on the target object; feature information of the target object is obtained; the position of the target object is monitored according to the feature information; when the situation that the position of the target object changes is monitored, focusing is performed on the target object again according to the feature information; when a shooting request is received, shooting is performed on the target object. The embodiment of the invention further discloses a shooting device. By the adoption of the shooting method and the shooting device, a clearer image can be obtained when a moving object is shot, a certain anti-shake effect is achieved, and the shooting operation is simplified.

The invention provides a synchronizer self-learning identification control method and a vehicle synchronizer position verification control method based on the synchronizer self-learning identification control method. According to the synchronizer self-learning identification control method, forward and reverse rotation of a gear selection and shifting motor are controlled, the voltage change of a gear selection and shifting position sensor is monitored so as to identify the state of a synchronizer reaching extreme positions, the voltage value of the position sensor when the synchronizer is at each extreme position is recorded, and therefore self-learning identification of gear selection information and gear shifting information of the synchronizer is realized. According to the position verification control method, when a speed changer reaches a bottom line, a vehicle runs for a certain period of time and the vehicle cannot be started normally due to gear selection and shifting, the synchronizer self-learning identification control method is adopted for learning and identification of the gear selection information and the gear shifting information. By means of the synchronizer self-learning identification control method and the position verification control method, the problem that the synchronizer is inaccurate in position because of multiple factors is effectively eliminated, gear selection and shifting of the vehicle can be based on a correct synchronizer position, and therefore the correctness of gear selection and shifting of the vehicle is guaranteed.

A device includes a character-data rotating section that rotates a regular-position character by a predetermined angle with respect to a reference point that is the center point of the background area of the regular-position character by using regular-position character data having a rotation angle of 0° and a center-point matching processing section that horizontally and / or vertically enlarges the background area of the rotated character data to cause the center point of the rotated character and the center point of BMP data to match each other even with respect to rotated character data. Thus, when multiple pieces of character data are arranged so that the center points thereof lie on a reference line, not only are the center points of the characters aligned along the reference line, but also bottom portions of the characters aligned with respect to the reference line.

The invention relates to an animal state monitoring method and device, electronic equipment and a storage medium, and the method comprises the steps of obtaining a breeding region image which comprises at least two animals; segmenting animals in the breeding area image through a pre-trained animal segmentation model to obtain animal aggregation information; and determining an animal state according to the animal aggregation information. The technical scheme is based on a computer vision mode; animals are segmented from images through a pre-trained model, animal aggregation information is obtained through analysis, and the animal state is determined according to the animal aggregation information, so that automatic monitoring of the animal state is achieved, abnormal conditions of the animals can be found timely and accurately; animal epidemic diseases are prevented, animal health is guaranteed, and breeding benefits are improved. In addition, manual monitoring is not needed, and laborcost and time cost are reduced.

The invention relates to the technical field of face recognition, in particular to a face snapshot gunlock load balancing method, which comprises the steps that two paths of original images are acquired, and an acquisition module of a snapshot gunlock transmits the original images to an image encoding and decoding module; an image encoding and decoding module outputs one path to a CPU encoding module for encoding, and outputs the other path to the GPU module for face detection; after receiving the image, a CPU image application module stores the image to a local storage card and uploads the image to a far-end server through a network; a CPU image analysis module analyzes the locally stored images, counts the number N of face snapshots in a set time and transmits a code rate value X needingto be adjusted, and the CPU encoding module acquires and configures an initial code rate value as the adjusted code rate value X; the coding code rate is dynamically adjusted in real time, the code rate is reduced when the snapshot amount is large, and the effects of giving way to CPU resources, locally storing pictures and uploading the pictures to a remote server are achieved; on the contrary,the code rate can be recovered and even improved in the time period with small snapshot amount, and a load balancing strategy is achieved.

The invention discloses a target detection method based on a YOLO-Terse network. The method comprises the steps of obtaining a to-be-detected image containing a to-be-detected target; inputting the to-be-detected image into a pre-trained YOLO-Terse network, and determining the category to which the to-be-detected target belongs and the position of the to-be-detected target in the to-be-detected image according to the features of the to-be-detected image, wherein the YOLO-Terse network is formed by adopting hierarchical and channel-level pruning on the basis of the YOLOv3 network and guiding the network to recover by combining knowledge distillation. According to the invention, layer pruning, sparse training, channel pruning and knowledge distillation processing are carried out on the YOLOv3, optimized processing parameters are selected, the simplified YOLO-Terse network is obtained, the size of the network is greatly reduced, most redundant calculation is eliminated, the target detection speed based on the network is greatly increased, and the detection precision can be maintained.

The invention relates to a teacher model classification layer weight-based output regularization method, which comprises the following steps of: converting the weight of a classification layer of a teacher model subjected to supervised training into a correlation matrix among categories, and taking each row in the matrix as a soft label of a corresponding category to provide additional information for a student model and participate in training of the student model; and selecting the student model with the highest accuracy as a final target model. According to the method, the information provided by the teacher model is fully utilized, the problems that the teacher model occupies too large training resources and the overall training time is too long in the training process are reduced, even if some neural network models can only provide the weight of a teacher model classifier layer, the student model can be trained through the method, and the training efficiency is improved. Compared with the prior art, the method has higher classification accuracy and wider model applicability, has higher training speed, only needs fewer training resources, and can further regularize the network model under the condition of fewer resources.

According to an aspect, there is provided a method for monitoring quality of loose tube fiber optic cable during manufacture in a secondary coating line. Initially, a trained machine-learning algorithm for calculating expected values of one or more quality metrics of manufactured loose tube fiber optic cable based on values of the one or more production process parameters of the secondary coating line is maintained in a machine-learning database. A computing system monitors one or more values of the one or more production process parameters during miming of the secondary coating line and calculates, in real-time during the monitoring, one or more expected values of the one or more quality metrics using the trained machine-learning algorithm with the monitored values of the one or more production process parameters as input. The computing system outputs at least the one or more expected values of the one or more quality metrics to a user device.

An information processing device includes: a width dimension acquisition section that acquires a width dimension of a tape on which a character image is to be printed; and an alternative setting section that sets a maximum alternative, which represents a selectable alternative having a maximum size among a plurality of alternatives for selecting a point size of the character image, according to the acquired width dimension so that the character image does not protrude from the tape.

A system, method and computer program product for identifying spam in an image, including (a) identifying a plurality of contours in the image, the contours corresponding to probable symbols; (b) ignoring contours that are too small or too large; (c) identifying text lines in the image, based on the remaining contours; (d) parsing the text lines into words; (e) ignoring words that are too short or too long from the identified text lines; (f) ignoring text lines that are too short; (g) verifying that the image contains text by comparing a number of pixels of a symbol color within remaining contours to a total number of pixels of the symbol color in the image, and that there is at least one text line after filtration; and (h) if the image contains text, rendering a spam / no spam verdict based on a contour representation of the text that which appears after step (f).

The invention relates to a training system for identifying defects of a display panel. The training system comprises an input module, a processing module, a generating module and a display module. Theinput module outputs an input signal according to user operation. The processing module is connected with the input module and generates identification information according to the input signal. Thegenerating module is connected with the processing module and provides defect information for the processing module. The display module is connected with the processing module and the generating module and displays a defect image which corresponds to the defect information. The processing module compares the identification information with the defect information and obtains a judgment result. Dueto the training system, the limit of real object training can be broken, practice opportunities for operators embarking on display panel defect identification can be increased, and the productivity can be further improved.