48 results about "Vehicle category" patented technology

The invention provides a vehicle detection method fusing adaptive attention in a congestion scene, and relates to the technical field of vehicle detection. For the problems of low vehicle detection precision and high omission ratio in a traffic jam scene, a YOLOv5 network model is constructed, an attention module is constructed based on a CBAM attention module and an ECA attention module, and the attention module is introduced into a backbone network of the YOLOv5 network model; based on the Wise-IoUv3 loss function and the Focal loss function, constructing a joint loss function, and replacing the loss function of the YOLOv5 network model with the joint loss function; respectively embedding ConvMixer hybrid convolution modules into three prediction branches of a head network of the YOLOv5 network model; the method comprises the following steps: detecting congestion vehicles at a monitoring view angle by using a trained model weight, obtaining vehicle category information, enhancing vehicle key feature extraction capability and multi-scale feature expression capability, optimizing bounding box positioning precision through a dynamic weighting mechanism, and promoting attention to a difficult-to-detect target.

The application relates to the field of warehouse management, in particular to a logistics warehouse management system based on big data, which comprises a division unit, a analysis unit and a control unit. The division unit is used for determining effective task points according to task update frequency, determining regional division modes according to the number and distribution coefficient of the effective task points, and determining the vehicle category of the carrying equipment according to the running efficiency and loading proportion. The analysis unit is used for determining analysis processing modes according to the vehicle matching degree of the carrying area. The control unit is used for determining the priority coefficient of each adjustable vehicle, determining the response vehicle according to the priority coefficient, and determining the moving route of the response vehicle according to the radiation demand degree of the movable route. The scheduling unit is used for determining the area state according to the task dispersion degree of the two types of carrying areas and the number of the response vehicles during vehicle scheduling analysis, determining the scheduling mode according to the area state, and determining the vehicle scheduling range according to the task fluctuation coefficient and the obstacle complexity. The logistics warehouse management efficiency is improved.

The invention provides a multi-modal fusion unmanned aerial vehicle identification method and system, and the method comprises the steps: obtaining a radio signal, an image signal and a sound signal of an unmanned aerial vehicle; performing feature extraction on the radio signal based on a blind source separation algorithm and adaptive filtering to obtain radio features; performing feature extraction on the image signal based on image super-resolution reconstruction and a double-flow feature extraction network to obtain image features; performing feature extraction on the sound signal based on spectral analysis and harmonic feature extraction to obtain an audio feature; fusing the radio feature, the image feature and the audio feature to form a multi-modal fusion feature; and performing identification based on the multi-modal fusion features to obtain an unmanned aerial vehicle category identification result. By adopting the scheme, the accuracy, reliability and environmental adaptability of unmanned aerial vehicle identification can be remarkably improved.

The invention provides an unmanned aerial vehicle identification and tracking method and system based on multi-modal fusion and trajectory modeling, and relates to the technical field of target detection and tracking. Infrared / visible light images and audio signals are synchronously collected in an airport key area, visual and acoustic features are extracted through a deep network after time alignment, and the target detection and tracking accuracy is improved. Outputting respective unmanned aerial vehicle category probabilities and candidate target frames; afterwards, weights are distributed adaptively according to confidence coefficients of audio and visual classification results, category fusion is carried out on a probability level, and unified multi-modal fusion representation is constructed on a feature level, so that relatively high recognition reliability is still kept in a small-target, weak-texture and strong-noise scene; on the basis, the multi-modal fusion feature sequence serves as input, the position and speed of the unmanned aerial vehicle in each time step are constructed into a trajectory state sequence, a sequence generation model outputs a complete trajectory in an autoregressive mode under conditional constraints, and continuous tracking, occlusion interval complementation and future trend prediction of the motion of the unmanned aerial vehicle are achieved.

A computer architecture for interactive image based database searching includes a first server that acquires category information indicating a category of a vehicle owned by a user. The first server also acquires part information for identifying a vehicle part located at a position specified by a user operation within an image of a vehicle displayed on a terminal device of the user. A second server is in communication with the first server and searches for products by accessing a storage storing thereon product information on vehicle part products and by using the acquired category information and the acquired part information as search criteria. The system transmits search result information indicating results of the search to the terminal device.

The invention relates to a method for operating an autonomous vehicle (1), in which the surroundings of the vehicle (1) and objects located in the surroundings are detected by means of detection signals from a surroundings sensor (3). According to the invention, the critical degree of the current situation detected by means of the signal behind the vehicle (1) is classified as a function of the determined vehicle class of the following vehicle (2) following the vehicle (1) and the detected distance (a) between the vehicle (1) and the following vehicle (2).

The techniques described herein relate to joint space / time domain beam management, including spatial domain beam management over an observation window and time domain beam management over a prediction window. Incorporating spatial and temporal dependencies increases the prediction interval and makes the prediction more accurate because spatial beam management is performed over the observation interval. Interaction aware multi-user equipment (UE) beam management techniques are also described that model a plurality of user interactions in a beam management procedure to predict future trajectory data and to predict an optimal future beam and / or a future optimal subset of beams per UE. For predicted future trajectory data and / or beam prediction data, a single user category or multiple user categories (e.g., vehicle category users, pedestrian category users) may be considered.

Image acquisition system (2000) for a traffic control automated system (1001) configured to: (a) measure (3001) the trajectory (TRJ), position (POS) and speed (VIT) of at least one vehicle (1003) using a road vehicle tracking device (2003); (b) illuminate (3002) the vehicle (1003) using an infrared illumination device (2001); (c) acquire (3003), using an infrared-sensitive camera (2002), at least one infrared image (IM-IR) of the vehicle (1003) illuminated by the infrared illumination device (2001); (d) process (3004) the infrared image (IM-IR) of said vehicle (1003) using a data processing device (2006); when the vehicle's speed (VIT), vehicle position (POS), vehicle category (CAT), and / or driver conduct (COMP) meets at least one offence criterion (CR-INF): (e) illuminating (3005) the vehicle with a visible light lighting device;(f) acquire (3006) a colour image (IM-VIS) of the vehicle (1003) illuminated using a colour photographic device (2005);

Disclosed herein are methods and systems for operating automated vehicles to accomplish an agricultural mission according to a hierarchically computed mission plan, comprising receiving a plurality of mission parameters defining an agricultural related mission, computing a generic plan for executing the mission by one or more of a plurality of vehicle classes grouping automated vehicles sharing common vehicle parameters coupled with one or more of a plurality of payload classes grouping payloads sharing common payload parameters, computing a plurality of specific plans for executing the mission by one or more automated vehicles of the selected vehicle class(s) coupled with one or more payloads of the selected payload class(s), selecting one of the plurality of specific plans which reduces one or more mission attributes of the mission and outputting instructions for operating the selected automated vehicle(s) and the selected payload(s) to execute the mission according to the selected specific plan.

System (10) for route planning of a vehicle (12) based on the road characteristics, wherein the system (10) comprises: a global navigation satellite system (GNSS) (16) for determining a geographic location of the vehicle (12); a vehicle communication system (20) for communicating with server systems (28); and a vehicle controller (14) in electrical connection with the GNSS (16) and the vehicle communication system (20), wherein the vehicle controller (14) is programmed: to determine a vehicle mass, a vehicle height and a vehicle width; to receive a navigation request from an occupant of the vehicle (12); and to determine a navigation route using at least the GNSS (16) and the vehicle communication system (20) in response to receiving the navigation request, wherein the navigation route is based at least partially on the vehicle mass, vehicle height and vehicle width, wherein, in order to determine the navigation route, the vehicle controller (14) is further programmed as follows: to determine the geographical location of the vehicle (12) using the GNSS (16); to transmit the vehicle's geographical location (12), navigation request, vehicle mass, vehicle height and vehicle width to a server controller (30); and to receive the navigation route from the server controller (30) using the server controller (30), which is programmed as follows: to determine several possible routes at least partially based on the geographical location of the vehicle (12), each of which satisfies the navigation requirement; to determine a route-drive comfort score for each of the several possible routes; and to select the navigation route from the several possible routes at least partially based on the route drive comfort score for each of the several possible routes, wherein the server controller (30) is electrically connected to a database, wherein the database contains several segment drive comfort records, wherein each of the several segment drive comfort records corresponds to one of several road segments of several lanes, and wherein each of the several segment drive comfort records has a location attribute, a vehicle class attribute and a drive comfort score attribute, wherein, in order to determine the route drive comfort score for each of the several possible routes, the server controller (30) is further programmed as follows: to determine a vehicle class of the vehicle (12) at least partially based on the vehicle mass, vehicle height and vehicle width; to retrieve a subset of the multiple segment drive comfort records from the database, wherein the location of each subset of the multiple segment drive comfort records is along one of the multiple possible routes, and wherein the vehicle class of each subset of the multiple segment drive comfort records corresponds to the vehicle class of vehicle (12); and The route drive comfort score for each of the multiple possible routes is determined by summing the drive comfort score of each subset of the multiple segment drive comfort data records along each of the multiple possible routes.

A method in a telematics system is provided. The method is for identifying an asset category of a plurality of asset categories. The method includes executing category identifying asset data queries after excluding error causing asset data requests from a main list and excluding asset categories from the main list based on the result of executing category identifying vehicle data queries. The steps are repeated until a single asset category remains in the main list.

The invention discloses an unmanned aerial vehicle battle scoring platform, a scoring mode and an unmanned aerial vehicle battle assembly. The unmanned aerial vehicle battle scoring platform comprises a bearing platform, a sensing module, an identification module and a statistics module, the bearing platform has a bearing surface; the sensing module is used for sensing whether the unmanned aerial vehicle is parked in the bearing surface; the identification module is used for identifying the type of the unmanned aerial vehicle parked in the bearing surface; the statistical module is connected with the sensing module and the recognition module, and the statistical module is used for performing timing or scoring statistics on the unmanned aerial vehicle parked in the bearing surface according to the sensing structure of the sensing module and the recognition result of the recognition module. The unmanned aerial vehicle battle scoring platform provided by the invention can improve the accuracy of unmanned aerial vehicle battle result statistics.

In some implementations, a classification system may define multiple vehicle segments that each include a set of vehicle makes based on a set of vehicle make and model combinations. The classification system may define, based on the set of vehicle make and model combinations, multiple vehicle categories based on subsets of the vehicle make and model combinations with similar attributes, wherein the multiple vehicle categories are each associated with a unique identifier and a respective vehicle segment, of the multiple vehicle segments. The classification system may define multiple vehicle indexes that are each associated with a respective modeling profile. The classification system may store information that associates each of the multiple vehicle indexes with one or more of the multiple vehicle categories.

A method for fine-grained traffic noise prediction includes obtaining location data indicating one or more locations and, for each respective location of the one or more locations, location-specific traffic data. The method includes, for each respective location of the one or more locations, using various computational models to predict a vehicle count and a vehicle class mix for the respective location; calculating, based on the predicted vehicle count and the predicted vehicle class mix, a number of vehicles per vehicle class for the respective location; and calculating, based on the number of vehicles per vehicle class, a predicted noise level for the respective location. The method includes causing a client device to display a map. The map may include, for each location of the one or more locations, a visual indication of the predicted noise level for the respective location.

This invention relates to the field of port vehicle management technology and discloses a hybrid positioning system for classifying and controlling port-owned and externally-entering vehicles. The system includes a vehicle classification and determination module, a multimodal location credential node, a module for actively binding and positioning owned vehicles, a module for passively binding location credentials for externally-entering vehicles, and a classification and control processing module. The multimodal location credential node collects optical response information, magnetic field response information, and millimeter-wave response information generated by the vehicle when it passes through the node's coverage area and generates location credential data. Externally-entering vehicles generate multimodal location response information through the passive location credential module to form location credential data. Vehicle positioning records or area positioning records are generated according to the vehicle category. This invention, by setting up multimodal location credential nodes and a passive location credential module for externally-entering vehicles, achieves the generation of area positioning records based on node passage records, reducing the loss of positioning information caused by differences in terminal installation conditions.

This invention discloses a method and system for constructing a high spatiotemporal resolution motor vehicle pollutant emission inventory, relating to the field of motor vehicle pollution monitoring technology. The method includes: acquiring vehicle driving information and regional electronic road network linear element files, preprocessing them to obtain a standardized motor vehicle activity dataset and a standardized road network file; matching vehicle trajectories with the regional road network to obtain the vehicle's travel volume on each road segment; determining the vehicle category on the road segment, acquiring historical travel volume data corresponding to the current category of vehicles, finding the historical travel volume most similar to the current category of vehicles from the historical travel volume data, and using the instantaneous pollutant emission rate corresponding to the historical travel volume as the instantaneous pollutant emission rate corresponding to the travel volume; and weighting the instantaneous pollutant emission rates of the road segment according to road length, road grade, and operational complexity to generate an emission inventory with kilometer-level spatial resolution.

The invention discloses a traffic tunnel pollutant time reduction factor calculation method, and belongs to the technical field of tunnel ventilation design. A reference year and a target year are set, a multi-dimensional vehicle classification system is established, the vehicle inventory, the driving mileage and emission data of the reference year are obtained, and a dynamic pollutant emission model and a vehicle market evolution model are constructed to accurately predict the inventory and the driving mileage of each classification vehicle in the target year. A dynamic pollutant discharge model is used for calculating the comprehensive pollutant discharge amount of a reference year and a target year, so that a pollutant time reduction factor is calculated, and a parameter self-adaptive adjustment mechanism is carried out by comparing prediction and actual monitoring data, so that key parameters of the model are automatically optimized. According to the scheme, a high-precision pollutant time reduction factor calculation method is provided, the precision of traffic tunnel pollutant emission prediction is remarkably improved, dynamic prediction of the vehicle ownership structure is achieved, and a scientific basis is provided for refined design and long-term planning of a tunnel ventilation system.

This invention discloses a reliable carbon activity measurement and credit issuance scheme for vehicle carbon rights confirmation. The onboard terminal collects one or more carbon activity-related data points according to time or mileage windows, including mileage, fuel consumption, estimated fuel injection, engine operating status, battery state of charge, charging amount, discharging amount, regenerative energy, or idling time. A carbon activity window summary is generated and bound to vehicle category identifier, metering version, emission factor version, monotonic counter reference information, and chain reference information, and then integrity protection is performed to form a carbon activity certificate. The server first performs signature verification, continuity verification, and joint review of vehicle category adaptation, metering version validity, and emission factor version availability on the carbon activity certificate. Then, based on the vehicle category identifier, metering version, and emission factor version, carbon rights conversion is performed. Credit confirmation receipts are issued only when the reliability threshold is met and there is no duplicate confirmation, thereby suppressing forgery, duplicate applications, and offline re-transmission of duplicate issuances, and improving the reproducibility and auditability of the conversion results.

The embodiment of the application discloses a vehicle re-identification model construction method based on multi-image learning, comprising: acquiring a vehicle image set, each vehicle image comprising a single vehicle and being labeled with a vehicle category; according to the space-time information of each vehicle image, combining multiple vehicle images taken at the same space-time position into an image package, and taking the vehicle category with the most labels in the image package as the labeled category of the image package; and training a vehicle re-identification model based on deep learning by using each image package. The embodiment improves the accuracy of vehicle re-identification.

The present invention provides a vehicle class data generation device, etc., that generates vehicle class data indicating the vehicle class that a driver can drive, taking into account the restrictions attached to the license type and their removal. [Solution] The vehicle class data generation device 3 includes: an acquisition means 310 that acquires license type information stored in the IC chip of a driver's license, restriction information indicating the restrictions attached to the license type by restriction text, and release information indicating the removal of restrictions by release text; a restriction release means 320 that extracts vehicle class information corresponding to the restriction text acquired by the acquisition means 310 based on a vehicle class table 300 which associates the restriction text with vehicle class information indicating the vehicle class that can be driven by the restriction text, and deletes the vehicle class information according to the release text acquired by the acquisition means 310; and a vehicle class data generation means 330 that uses the license type information and the vehicle class information that remains without being deleted by the restriction release means 320 to generate vehicle class data indicating the vehicle class that the holder of the driver's license can drive.

The invention discloses a traffic key vehicle determination method fusing visual attributes and OBU registration information, and relates to the technical field of traffic supervision, and the method comprises the steps: collecting vehicle visual features through a high-definition checkpoint camera at a traffic node entrance, and generating a first vehicle attribute set; vehicle official registration data is obtained using wireless communication between a roadside unit (RSU) and an on-board unit (OBU) to generate a second set of vehicle attributes. After license plate consistency verification is completed and data alignment is realized, differential credibility empowerment is performed on two types of attribute sets, the high authority of OBU registration information and the characteristic that a visual identification result is influenced by the environment are fully considered, and finally, multi-source data fusion judgment is completed through a preset rule engine, and a vehicle type judgment result is accurately output. And linkage of local storage and platform data is realized.

This application discloses a method and apparatus for identifying vehicles, a non-volatile storage medium, and an electronic device. The method includes: acquiring an image to be identified, wherein the image contains the vehicle to be identified and its driving environment; processing and analyzing the image using a vehicle detection model to obtain analysis results of vehicle detection output, wherein the analysis results include: the vehicle category and vehicle location of the vehicle to be identified in each image; the vehicle detection model undergoes multi-stage training, which includes: training the vehicle detection model to extract domain-invariant features based on the distribution differences between the source and target domain images at multiple feature scales, including a semantic scale; and domain-invariant features are image features where the distribution difference between the source and target domain images at the semantic scale is less than a preset threshold.

This application relates to a vehicle control method, device, controller, vehicle, and storage medium. The method includes: acquiring cutting-in intention information of a second vehicle in an adjacent lane next to the lane occupied by a first vehicle; the cutting-in intention information is determined based on vehicle information monitored for the second vehicle, and includes cutting-in intention intensity and cutting-in probability; determining an intention intensity range based on a preset intention intensity threshold; determining the cutting-in intention type corresponding to the intention intensity range as the cutting-in intention type of the second vehicle if a preset type determination condition corresponding to the intention intensity range is met based on vehicle information; determining a control strategy for the first vehicle based on the cutting-in intention type, cutting-in intention intensity, cutting-in probability, and the vehicle category of the second vehicle; and controlling the first vehicle to drive according to the control strategy. This method can effectively balance the intensity of the cutting-in game with vehicle driving safety.

This application provides a method and apparatus for detecting vehicles driving in the wrong direction on highways. The method acquires monitoring images of vehicles on the highway, inputs these images into an image detection model, tracks vehicle bounding box information based on vehicle category confidence, obtains vehicle trajectory data, determines vehicle motion vectors based on the trajectory data, samples regions of interest within lanes, inputs the coordinates of the sampled feature points into a motion vector estimation model to obtain displacement vectors of the sampled feature points, clusters the displacement vectors of the sampled feature points based on their reachability distance to obtain target clusters, uses the mean of the displacement vectors in the target clusters as background displacement vectors, and determines the vehicle driving in the wrong direction detection result based on the vehicle motion vector and the background displacement vector. This application improves the accuracy of vehicle driving in the wrong direction detection result and meets the needs for automatic real-time monitoring, identification, tracking, and detection of vehicle driving in the wrong direction in highway scenarios.

A system and method of calibrating a sensor (15) of a vehicle (12) includes providing a plurality of targets (14) for calibrating a sensor (15) on an equipped vehicle (12) of a plurality of vehicle categories, where each vehicle category defines a position of at least one sensor (15) on a vehicle (12) of the vehicle category, the method further includes orienting a vehicle (12) at a reference position on a vehicle platform (22), and moving the vehicle (12) based on a vehicle category of the orientated vehicle to align the plurality of targets (14) with the sensor (15) for calibrating the sensor (15).

This invention relates to the field of electric vehicle control technology, and provides a method and system for complex road condition intention recognition and decision-making in intelligent driving of electric vehicles. The method includes: acquiring vehicle driving information, preceding vehicle point cloud, and image data using onboard radar and cameras based on a unified coordinate system to ensure spatial alignment of perception data; selecting the closest point to the vehicle from the preceding vehicle point cloud as a benchmark and constructing a horizontal plane to simplify the three-dimensional point cloud into two-dimensional data, eliminating redundant information and reducing computational load; classifying the target point cloud into three categories according to the spatial distribution direction to accurately match the relative positions of the preceding vehicle and the vehicle; adapting and fitting different point cloud sets to generate preceding vehicle boundaries, i.e., warning lines; detecting the position and deflection angle of the warning lines in real time, and determining the preceding vehicle's lane-changing intention by combining the included angle, turn signal, and offset; and controlling the vehicle to decelerate or brake according to the preceding vehicle category, thereby improving the decision-making response speed and driving safety of intelligent driving in complex road conditions.

Methods (100, 200, 300, 400, 700, 800) for controlling a carrier vehicle (10), comprising: Acquisition (102) of objects (RO) by means of a control (34) of the carrier vehicle (10), wherein the acquisition (102) of objects (RO) includes receiving remote object data, which includes remote object data class data and location data, and which includes remote vehicles (RV) and remote infrastructure (RI); Assign (104), by the control (34), an object class to each of the detected objects (RO) based on the class data; Assign (106), by the control (34), a priority value for each of the captured objects (RO) based on the object class of each of the captured objects (RO); Assigning (108), by the controller (34), a transmit time interval, TTI, to cooperative sample messages, CSMs, based on the priority value, to modify the TTI of each of the CSMs according to the assigned TTI; Transfer (109) the CSMs to the remote vehicles (RV); where the object class includes an unprotected road user, VRU, class, a vehicle class, a traffic sign class, and another class; where the vehicle class refers to remote vehicles (RV); where the VRU class refers to pedestrians and cyclists; and where the traffic sign class refers to traffic signs and traffic symbols; wherein the assignment (104), by the control (34), of the object class to each of the captured objects (RO) based on the class data includes the assignment (104) of the VRU class to at least one of the remote objects (RO) in response to the capture (102) of the objects (RO) to indicate that at least one of the remote objects (RO) is a VRU; where the assignment (106), by control (34), of the priority value to each of the captured objects (RO) based on the object class of each of the captured objects (RO) includes: Assigning (106, 202, 204) a VRU priority value in response to assigning (106, 204) the VRU class to at least one of the remote objects (RO); Determine (304), based on the location data, that the VRU is moving; Increasing (304) the VRU priority value by an initial, predetermined VRU value in response to determining (304) that the VRU is moving; Determine (306), based on the location data, whether the VRU is located on a sidewalk (CW); Increasing (306) the VRU priority value by a second, predetermined VRU value in response to determining (306) that the VRU is on the sidewalk (CW); and where a change in the VRU priority value is equal to the sum of the first, predetermined VRU value and the second, predetermined VRU value.

The invention discloses a vehicle classification method and device, a storage medium and computer equipment, and relates to the technical field of financial science and technology. The method is mainly used for improving the classification efficiency and the classification precision of vehicles. Comprising the following steps: in response to a vehicle classification signal of a to-be-classified vehicle, judging whether a target vehicle matched with the to-be-classified vehicle exists in a preset vehicle classification library or not, if so, determining the category of the to-be-classified vehicle based on the category of the target vehicle, otherwise, collecting a tire multispectral image and a three-dimensional point cloud of the to-be-classified vehicle, the preset vehicle classification library is updated in real time based on the new vehicle category; extracting multispectral tire feature data of the to-be-classified vehicle from the tire multispectral image, and determining vehicle body geometric parameters of the to-be-classified vehicle based on the three-dimensional point cloud; and inputting the multispectral tire feature data and the vehicle body geometric parameters into a preset classification model for vehicle classification to obtain the category of the to-be-classified vehicle. The method is mainly suitable for vehicle classification scenes.

A method in a telematics system is provided. The method is for identifying an asset category of a plurality of asset categories. The method includes executing category identifying asset data queries after excluding error causing asset data requests from a main list and excluding asset categories from the main list based on the result of executing category identifying asset data queries. The steps are repeated until a single asset category remains in the main list. A telematics system including a telematics server, a telematics database coupled to the telematics server, a network, and a telematics device is provided, the telematics system configured for carrying out the method.

The invention provides a vehicle category detection method and system, terminal equipment and a storage medium. The method comprises the steps of determining a target driving vehicle and a vehicle driving area; obtaining regional obstacle vehicles, a regional obstacle vehicle category detection frame and regional obstacle vehicle point cloud data based on the target driving vehicle and the vehicle driving region; obtaining regional obstacle vehicle two-dimensional data and regional obstacle vehicle overlook data based on the regional obstacle vehicle point cloud data; obtaining a barrier vehicle matching result based on the regional barrier vehicle category detection frame, the regional barrier vehicle two-dimensional data and the regional barrier vehicle overlook number, and dividing the regional barrier vehicles into target barrier vehicles and candidate barrier vehicles; and obtaining a candidate vehicle category based on the candidate obstacle vehicles, the regional obstacle vehicle two-dimensional data, the regional obstacle vehicle category detection frame, the regional obstacle vehicle overlook data and a preset weight mapping relationship, and further obtaining the regional obstacle vehicle category, so as to improve the vehicle classification accuracy.