53 results about "Curve fitting algorithm" patented technology

The invention discloses a thyroid cancer analysis method and device based on real-time fluorescent quantitative PCR. The method comprises the following steps: acquiring to-be-analyzed data; determining sample exclusive threshold line data according to the to-be-analyzed data; calculating a corrected Ct value based on double inflection point dynamics according to the exclusive threshold line data of the sample so as to obtain a final corrected Ct value; and generating a detection result according to the final corrected Ct value. Through optimization processing of the curve fitting algorithm, the linear characteristic of the fluorescence amplification curve is remarkably improved, the interference of background noise on the result is reduced, and the calculation stability of the Ct value is improved.

The invention discloses a logging imaging fracture-cavity segmentation and parameter extraction method based on deep learning. The method comprises the following steps: step 1, making a data set; step 2, model training and selection: inputting the data set made in the step 1 into a deep learning model for training, and generating two key outputs, namely last. Pt and best.pt, by the model in the training process; best.pt is selected as a model in the actual segmentation process; 3, further designing a client interface of the logging imaging crack segmentation and parameter extraction system on the basis of completing data set production and model training and selection; step 4, crack image processing and curve fitting algorithm implementation, and crack inclination angle parameter extraction; and 5, opening a client interface, selecting a data source, starting detection, outputting a result, and displaying the result. According to the invention, crack inclination angle parameter information can be accurately extracted; and an integrated and visual operation platform is provided for a user.

Some embodiments of the specification provide a welding method based on coordinate system registration, the method comprising: acquiring, by a sensor, a sequence of spatial coordinate points of a teaching welding path and teaching point cloud data of a standard workpiece; generating, by a curve fitting algorithm, a standard welding path based on the sequence of spatial coordinate points; determining a reference point based on the teaching point cloud data and the sequence of spatial coordinate points; constructing a welding database based on a first workpiece identifier of the standard workpiece, the reference point, and the standard welding path; acquiring a second workpiece identifier and coarse scanning point cloud data of a workpiece to be welded; acquiring, from the welding database, the teaching point cloud data, the reference point, and the standard welding path corresponding to the second workpiece identifier; determining a first registration transformation based on the teaching point cloud data and the coarse scanning point cloud data; determining a scanning range based on the reference point and the first registration transformation; acquiring fine scanning point cloud data based on the scanning range; and determining a welding instruction according to the fine scanning point cloud data and the standard welding path.

The invention relates to a railway track identification method and system based on yolov5 deep learning, and the method comprises the steps: inputting the image data information, collected in real time, of a railway track into a trained yolov5 deep learning detection model, and carrying out the identification and classification of a switch rail, a stock rail and a guard rail of the railway track, obtaining data information of a matrix of position points of a switch rail, a stock rail and a guard rail of the railway track; and fitting the contour lines of the switch rail, the stock rail and the guard rail of the railway track by adopting a global search-oriented adaptive weight nonlinear curve fitting algorithm to obtain data information of the fitted contour line set of the switch rail, the stock rail and the guard rail of the railway track. According to the method, the turnout of the railway track can be accurately distinguished and classified in combination with the yov5 deep learning detection model, manual participation is not needed in the whole process, manual errors are reduced, and the running stability and safety of a train are improved.

The invention provides an intelligent online correction method and system for pull-up, and relates to the technical field of data processing, and the method comprises the steps: obtaining real-time to-be-corrected pull-up video data; extracting skeleton point coordinate data in the pull-up video data through a posture recognition algorithm; missing value completion processing is carried out on the skeleton point coordinate data; establishing a joint state space model according to the skeleton point coordinate data after completion processing; according to the joint state space model, a target angle is predicted in combination with a Kalman filtering algorithm and a curve fitting algorithm; and outputting a correction prompt through a preset correction rule under the condition that the predicted target angle which does not conform to the pull-up motion standard index exists. According to the method and the system, a relatively good recognition and correction effect can be achieved on nonstandard actions in pull-up sports, correction feedback can be given to a user in real time according to a correction rule, and the user is assisted to better independently complete pull-up sports training.

This application discloses a temperature curve fitting algorithm, including the following steps: (1) preheating temperature control and feature extraction; (2) constant power curve fitting; (3) volume estimation; (4) anomaly handling. The temperature curve fitting algorithm of this application stabilizes to 50℃±3℃ within 10 minutes, and mines the temperature data features of the preheating stage through sliding residual trend analysis, capturing the weak heat capacity influence of small-volume insects in advance, and providing a sensitivity marker for the slope extraction in the subsequent constant power stage; the fitting mode is automatically switched by the slope reduction threshold, which not only retains the accurate extraction of the core slope in the stable segment, but also avoids the fitting distortion in the slowing segment; the fitting benchmark is dynamically corrected by h, rather than relying on a fixed insect-free curve, which effectively offsets the heat dissipation difference interference caused by room temperature fluctuations; when the heat dissipation coefficient estimation is abnormal, it is automatically downgraded to a linear fitting benchmark, and when historical data is insufficient, a backup formula is called to ensure that effective results can still be output under extreme scenarios such as sensor fluctuations and data loss.

The application discloses a curve segmentation fitting method and device for image segmentation and a storage medium, and comprises the following steps: pre-processing an image to obtain contour data; performing a contour segmentation algorithm of a DCE evaluation standard according to the contour data to obtain a segmentation point list; and performing a curve fitting algorithm based on an Euler curve model according to the segmentation list to obtain a final fitting result. The automatic segmentation of the contour can be realized, and subjective errors caused by human intervention are reduced.

This invention discloses a panoramic image monitoring system, comprising: an image acquisition module for acquiring visual image data through an ultra-wide-angle image acquisition device with at least four sets of ring-shaped first devices; a data processing module for sequentially correcting and stitching the acquired image data to obtain panoramic image data, then performing target detection, recognition, and tracking operations on the processed image, and employing linear interpolation or curve fitting algorithms to improve the detection and recognition accuracy of the target; and a display module for displaying the processed image data and the detection, recognition, and tracking results of the target. This invention solves the problem of potential obstruction of the camera's field of view or severe degradation of image quality, effectively improving the system's stability and reliability. It also enables precise tracking of object motion, improving the accuracy and continuity of dynamic target tracking, and facilitating effective handling of situations where multiple targets move simultaneously or change rapidly.

The application belongs to the technical field of path planning, and particularly relates to a PCB glue tearing machine operation path optimization control method and system, which comprises the following steps: collecting torque and motion state data of a glue tearing machine actuator, performing gravity compensation and filtering on the torque data; calculating accumulated mechanical work according to the processed torque data; obtaining a sticking resistance coefficient according to the accumulated mechanical work, displacement distance and torque change rate; when a continuous preset number of sticking resistance coefficients exceeds a threshold value, sampling control points on an initial path, taking the control point longitudinal coordinate as an optimization variable; constructing a cost function comprising a safety penalty term, a path smoothness penalty term and an efficiency reward term, performing particle swarm optimization on the control points according to the cost function to obtain optimal control points, and generating an avoidance path using a curve fitting algorithm on the optimal control points; and controlling the actuator to complete glue tearing according to the avoidance path. The application adjusts the path by monitoring the sticking state, thereby improving the glue tearing quality.

The invention relates to the technical field of atomizer control, in particular to an intelligent atomization control method and system for an atomizer, and the method comprises the steps: obtaining parameters of an equivalent model of a piezoelectric transducer; analyzing parameter fluctuation characteristics of the piezoelectric transducer under different external conditions according to parameters of the equivalent model; constructing an overall equivalent circuit of the atomizer, wherein the overall equivalent circuit comprises an LC filter and a piezoelectric transducer at the output end of an inverter; applying one or more pulse voltages to the overall equivalent circuit to trigger a transient phenomenon, measuring a transient current in the circuit, and determining the overall resonant frequency; based on the nonlinear correlation between the whole resonant frequency and the mechanical resonant frequency of the piezoelectric transducer, fitting related data by using a curve fitting algorithm; and according to the result of the curve fitting algorithm, calculating and tracking the mechanical resonant frequency of the piezoelectric transducer, and applying the tracked mechanical resonant frequency to the atomization system. The problems of slow tracking and low efficiency caused by fluctuation of the resonant frequency of the piezoelectric transducer in an existing atomization system due to external conditions can be solved.

The invention relates to the technical field of image recognition processing, and discloses a direct-injection English read-through card classification method based on image recognition processing, and the method comprises the steps: obtaining a damaged English read-through card image and a historical learning record of a user, recognizing a damaged region, and extracting the features of an individual forgetting mode; calculating a memory attenuation degree and a credibility coefficient based on an improved Ebbinghaus forgetting curve fitting algorithm; utilizing a memory guide content reconstruction algorithm, and taking the memory credibility as a constraint weight to adjust an image restoration parameter; semantic probability distribution is calculated based on the vocabulary features, and restoration candidate schemes are generated in combination with memory constraints; analyzing the matching degree between the candidate scheme and the user memory state by using a time sequence memory verification algorithm; verifying the semantic rationality of the repaired content based on a language model; and performing content identification on the restored image to generate a classification result of comprehensive restoration and time sequence optimization. According to the method, intelligent image restoration of semantic constraint is realized, and integrated processing of restoration and classification is realized.

The present application relates to a kind of plane short weld seam identification method, welding method and device, belong to the field of automated welding;Through the two-step photographing method of single camera autonomous welding model, accurately identify the three-dimensional coordinates of spatial point under the robot base coordinate system, give the autonomy of robot welding non-standard plane weld or batch plane weld;YOLOv5 neural network is used to locate the arc starting point, arc extinguishing point, workpiece corner point and weld area, improve the perception ability of robot to space target;Through brightness and contrast enhancement algorithm, edge detection algorithm, morphological processing algorithm, accurately extract weld point cloud from weld area;Through stereo matching algorithm and B spline curve fitting algorithm, identify and fit welding trajectory from irresistible, dynamic, irregular interference information;The fitted welding trajectory is transmitted to robot system to carry out autonomous welding on plane weld.

The invention provides an air conditioner which comprises an indoor unit, a heat exchange air duct and a fresh air duct are arranged in the indoor unit; the fresh air fan is arranged in the fresh air duct; the purification module is arranged in the heat exchange air duct; the controller is configured to calculate the range of the pollutant concentration sensed by the human body in the current sampling period by adopting an index curve fitting algorithm on the basis of the pollutant concentration in the current sampling period, the reference concentration and the upper limit value of the pollutant concentration in the previous sampling period; linearly converting the range of the human body sensing concentration into a preset numerical value range, wherein numerical values in the preset numerical value range are represented by s; dividing the preset numerical range into a plurality of continuous concentration intervals according to the boundary concentration; and according to the concentration interval where the current numerical value s is located, the fresh air fan and the purification module are controlled to operate according to the preset control logic. In the invention, the fresh air fan and the purification module are gradually adjusted according to the change of the human body perceived concentration, the intelligent level is high, and the user experience is favorably improved.

A kind of machining datum self-positioning method of additive repair bearing bush, comprising the following steps: S1. obtaining point cloud data of bearing bush surface;S2. preprocessing point cloud data;S3. point cloud registration aligns bearing bush surface with point cloud model;S4. after surface reconstruction, obtain feature points of bearing bush surface;S5. based on point cloud data, accurately extract feature points, and then calculate the geometric center of bearing bush;S6. determine the machining axis of bearing bush by curve fitting algorithm;S7. real-time acquisition of bearing bush positioning data determines machining datum and axis parameters, and automatically adjusts the position and parameters of machining equipment through compensation logic and three-dimensional reconstruction algorithm, so that the machining equipment matches the actual shape and position of bearing bush, so that the bearing bush is placed on the machining platform, the computer can automatically identify and determine the machining datum, axis and other parameters, and the position of machining equipment can be adjusted in real time through compensation method to match the bearing bush, improve the machining efficiency and machining precision, and have good practical value and economic value.

The application relates to a combined positioning method of a vehicle-mounted platform and a UAV, which comprises the following steps: acquiring LiDAR point cloud and vehicle coordinates from the vehicle-mounted platform, acquiring UAV coordinates and a first relative pose between the vehicle and the UAV from the UAV, generating a semantic map according to the LiDAR point cloud based on a parameterized curve fitting algorithm, determining accurate coordinates of the UAV according to the vehicle coordinates and the first relative pose, sending the semantic map to the UAV to instruct the UAV to shoot image data according to lane lines in the semantic map, obtaining a second relative pose between the UAV and the vehicle-mounted platform according to the image data and the semantic map, and determining accurate coordinates of the vehicle based on the second relative pose and the accurate coordinates of the UAV. Through the application, the problem that the vehicle and the UAV are not accurately positioned is solved, the bidirectional correction reduces the cumulative error, and the lightweight semantic map reduces the transmission bandwidth occupation while maintaining a high recognition rate.

The invention relates to a vehicle-mounted platform and unmanned aerial vehicle combined positioning method, and the method comprises the steps: obtaining a LiDAR point cloud and a vehicle coordinate from a vehicle-mounted platform, obtaining an unmanned aerial vehicle coordinate and a first relative pose between a vehicle and an unmanned aerial vehicle from an unmanned aerial vehicle, generating a semantic map according to the LiDAR point cloud based on a parameterized curve fitting algorithm, determining accurate coordinates of the unmanned aerial vehicle according to the vehicle coordinates and the first relative pose, sending a semantic map to the unmanned aerial vehicle to instruct the unmanned aerial vehicle to shoot according to lane lines in the semantic map to obtain image data, obtaining a second relative pose of the unmanned aerial vehicle and the vehicle-mounted platform according to the image data and the semantic map, and sending the second relative pose to the vehicle-mounted platform. And determining accurate coordinates of the vehicle based on the second relative pose and the accurate coordinates of the unmanned aerial vehicle. Through the method and the device, the problem of inaccurate positioning of the vehicle and the unmanned aerial vehicle is solved, accumulated errors are reduced through bidirectional correction, and the transmission bandwidth occupation is reduced while the lightweight semantic map keeps a high recognition rate.

The application discloses a flight pipeline modeling method for cluster unmanned aerial vehicle control, and establishes a geometric pipeline model of unmanned aerial vehicle flight; a profile model of a convex polygon section is established based on rotation angles and lengths of convex vectors, and the convex polygon section is composed of a plurality of convex vectors; nodes are constructed by positions of the convex polygon section on a reference curve, the rotation angles and the lengths of the convex vectors, the geometric pipeline is discretized into a plurality of node sequences, a curve fitting algorithm is applied to the node sequences to generate a continuous three-dimensional pipeline model; a comprehensive cost function of the geometric pipeline is established according to characteristics of the geometric pipeline; and the optimal geometric flight pipeline model is obtained when the comprehensive cost function is the minimum. The flight pipeline is modeled by the reference curve and the convex polygon section, a polygon geometric pipeline is obtained, and safe flight space modeling in a complex three-dimensional scene is realized.

The present application belongs to the technical field of digital measurement, and particularly relates to a high-precision three-dimensional measurement method for local features of an aircraft shape. First, the imaging center of a camera is aligned with a region to be measured on the surface of the aircraft, and the longitudinal position of a scanning system is adjusted so that the focal plane of the camera is located at an effective scanning position. Then, the included angle between the measurement light path and the surface to be measured is adjusted so that the number of imaging fringe periods is less than 1. The scanning system is controlled to move in a single direction to achieve longitudinal scanning of the surface to be measured. After scanning is completed, the longitudinal light intensity curve is extracted for each pixel point, and the Fourier transform technique and Gaussian curve fitting algorithm are further combined to obtain the accurate focusing position of the pixel point. Finally, height mapping is performed in combination with the scanning step distance to restore the three-dimensional topography of the region to be measured. The technical solution can improve the measurement accuracy, support three-dimensional measurement of small local features of an aircraft shape, and has the characteristics of high demodulation efficiency and wide adaptability.

The invention discloses a full-field deformation monitoring system and method for a cylindrical shell structure under hydrostatic pressure, and belongs to the field of deformation monitoring. Comprising the steps that strain sensors are arranged on the outer surface of equipment to be detected, strain of arrangement points is monitored, a radial strain continuous distribution diagram is drawn based on a cubic spline curve fitting algorithm, and the deformation condition of the cylindrical shell structure is monitored; and a pressure sensor is arranged at the hydraulic cabin to monitor the relationship between the strain and the pressure. According to the invention, full-field monitoring of the cylindrical shell can be realized, the strain and pressure change can be monitored in real time, and the method can be used for monitoring the real-time deformation of the surface of the cylindrical shell structure in an internal or external hydrostatic pressure environment.

Embodiments estimate the trajectory of an airborne or aquatic object. Embodiments comprise obtaining (204) track data of an airborne or aquatic object; fitting (210) at least one motion model to the track data using a curve- fitting algorithm to determine parameter values of the at least one motion model, and estimating (210) a trajectory of the object using the parameter values. The at least one motion model estimates the trajectory of the object as a plurality of linearly separable sections.

The present invention discloses a digital calibration method and system for a multi-phase clock, which relates to the technical fields of digital communication systems and digital signal processing technologies. The method includes: selecting a training sequence type and configuring a sampling circuit; fixing the PI Code, sampling the input data according to the set training sequence, recovering the corresponding clock phase of the initialized PI Code through a curve fitting algorithm, and sequentially fixing multiple PI Codes; analyzing the deviation between the clock phases of each phase obtained by actual measurement and the ideal phase, and quantifying the systematic deviation by plotting a curve; calculating a pre-distortion function according to the difference between the actual phase curve and the ideal phase curve; storing the calculated pre-distortion function in the non-volatile storage space of the chip; when the chip is working normally, reading the stored pre-distortion function and performing real-time compensation through a digital logic circuit to generate a calibrated clock phase index. The present invention realizes high-efficiency clock recovery and data sampling alignment through the digital calibration method and the pre-distortion function.

This invention discloses a novel continuum robot and its shape prediction method. The continuum robot includes a handle body, a snake skeleton, a drive device, and two drive cables. The snake skeleton is fixed to the front end of the handle body, and the drive device is located at the rear end inside the handle body. The two drive cables are symmetrically arranged axially within cable constraint grooves on both sides of the snake skeleton. The drive device controls the deformation of the snake skeleton through the drive cables. The snake skeleton includes several sequentially adjacent guide joints. Each guide joint includes a flexible hinge in the middle and a nested ring structure on each of the left and right sides that rotates around the flexible hinge. The shape prediction method utilizes the real-time collected changes in the length of the drive cables to obtain the predicted coordinates of several control points of the snake skeleton through a trained prediction model. The shape curve of the snake skeleton is reconstructed using a spline curve fitting algorithm. The continuum robot of this invention possesses good bending consistency, torsional resistance, and resistance to local disturbances, and exhibits high shape prediction accuracy.

The invention discloses a panoramic image monitoring system, and the system comprises an image collection module which is used for collecting visual image data through at least four groups of ultra-wide-angle image collection devices surrounding a first device; the data processing module is used for sequentially carrying out correction splicing processing on the collected image data to obtain panoramic image data, then carrying out specified target detection and recognition tracking operation on the processed image, and improving the detection and recognition precision of a specified target by adopting a linear interpolation or curve fitting algorithm; and the display module is used for displaying the processed image data and the detection and identification tracking result of the specified target. The method solves the problem that the visual field of the camera may be blocked or the image quality is severely reduced, can effectively improve the stability and reliability of the system, can achieve the precise tracking of the motion state of an object, improves the accuracy and continuity of dynamic target tracking, and improves the user experience. And simultaneous movement or rapid change of a plurality of targets can be conveniently and effectively handled.

The invention discloses a main transformer differential protection automatic calculation method and system, and belongs to the technical field of power system relay protection. Traditional main transformer differential protection calculation depends on manpower, is high in complexity, is easily influenced by parameter errors and is low in efficiency. Multi-source parameters such as a main transformer nameplate parameter, a CT / PT transformation ratio and a winding wiring mode are automatically acquired and verified through the parameter analysis module; self-adaptively calculating the balance coefficient of each side by using a dynamic balance coefficient calculation algorithm; a phase compensation matrix is automatically generated based on the wiring mode, and compensation effect verification is carried out; characteristic inflection points of the proportional braking characteristic curve are automatically searched through a six-phase current synchronous injection method, and an optimized proportional braking characteristic curve is generated through a curve fitting algorithm; and finally, carrying out optimization recommendation and dynamic adjustment on the differential fixed value by means of an AI auxiliary decision-making module. According to the method, full-process automation from parameter input to constant value generation is realized, the calculation efficiency and precision are remarkably improved, and rapid and accurate setting calculation support is provided for differential protection of the main transformer.

An apparatus, system, and method for compensating for bone loss of a patient's knee joint during the performance of an orthopaedic surgical procedure includes determining an estimated amount of bone loss of a bone of a patient's knee joint, determining an anatomical distance between a femoral anatomical feature of a patient's femur and a tibial anatomical feature of a patient's tibia, and adjusting the determined anatomical distance based on the estimated amount of bone loss. The determined anatomical distance is indicative of a ligament length of a collateral ligament of the patient's knee joint. The bone loss may be estimated based on, for example, a surface curve fitting algorithm applied to a three-dimensional model of the bone, a defined bone loss amount provided by a user, or a comparison of the three-dimensional model of the bone to a statistical shape model that approximates the bone in a healthy state.

The invention discloses a digital calibration method and system for a multi-phase clock, and relates to the technical field of digital communication systems and digital signal processing, and the method comprises the steps: selecting a training sequence type, and configuring a sampling circuit; pI Codes are fixed, input data are sampled according to a set training sequence, corresponding clock phases of the initialized PI Codes are recovered through a curve fitting algorithm, and a plurality of PI Codes are fixed in sequence; analyzing the deviation between the actually measured phase of each phase clock and the ideal phase, and quantifying the systematic deviation by drawing a curve; calculating a pre-distortion function according to the difference between the actual phase curve and the ideal phase curve; the pre-distortion function obtained through calculation is stored in a nonvolatile storage space of a chip; and when the chip works normally, the stored predistortion function is read, real-time compensation is carried out through the digital logic circuit, and a calibrated clock phase index is generated. According to the invention, high-efficiency clock recovery and data sampling alignment are realized through a digital calibration method and a pre-distortion function.

The application provides a laser radar aliasing waveform decomposition method based on waveform characteristics, constructs a waveform mathematical model according to seed light characteristics of a laser radar, combines an LM curve fitting algorithm, decomposes aliasing echoes into multiple single echo data, and achieves the purpose of accurately extracting multi-target ranging information. The technical basis of the application is a laser radar full waveform detection technology, that is, a laser radar detection system can sample analog electric signals converted from optical signals into digital waveform signals, and the processing object of the application is a waveform sampling point sequence of seed light and return light detected by the laser radar.

The invention relates to the technical field of image processing, in particular to a coastline automatic extraction method and system based on deep learning and satellite images. Comprising the steps of obtaining multi-temporal high-resolution image data of a target coast, and performing preprocessing to obtain an enhanced image sequence; extracting a connected component of the salinization erosion area, and generating a refined boundary pixel set; a distance field is generated, the minimum distance from each boundary pixel to the nearest boundary is calculated, an initial center point sequence is generated, and a spline curve fitting algorithm is adopted for optimization; generating a polygon buffer area, and performing superposition verification on the polygon buffer area and the enhanced image sequence to obtain a verified central point sequence; a continuous vector path is generated through fitting, coastline vector data are extracted, multi-temporal change monitoring is carried out, and high-salinity soil accumulation and erosion rate indexes are obtained; and updating the pixel-level classification to generate a final coastline vector data sequence. According to the method, the salinization erosion coastline can be efficiently and accurately extracted, and reliable data support is provided for ecological environment monitoring.

The invention discloses a microscopic visual detection method for surface defects of a nickel-titanium root canal file, and provides corresponding detection algorithms for three different defects of the root canal file. For the defect of metal peeling on the surface of the root canal file, a detection algorithm based on gray histogram comparison of OpenCV4 is provided; for the root canal filing spiral defect, a defect detection algorithm based on OpenCV4 curve fitting is provided; for the fatigue crack of the root canal file, a detection algorithm based on improved YOLOv8 is provided. The accuracy and feasibility of the algorithm are verified through experiments.

Aiming at the problems of wear and blockage in lane line detection, a lane line robust detection and extraction method based on road constraint is proposed. Firstly, a large number of Dongfeng park scene images are collected to construct a lane line dataset, and the road area is labeled while labeling the lane line. Secondly, the ENet-SAD network model is improved, and the lane line segmentation loss function of joint road constraint is defined to generate accurate lane line segmentation prediction results. Finally, the full connection conditional random field algorithm is introduced to post-process the output lane line probability map, correct the probability map, select N offline coordinate points on each lane line, and use the cubic spline curve fitting algorithm to obtain the fitted lane line. The experimental results show that compared with the ENet-SAD method, the method in this paper achieves good results in different scenes such as conventional, strong light, shadow, blockage, etc. At the same time, based on the loss function defined in this paper, the performance of existing lane line detection models (such as ENet-SAD, SCNN, etc.) can be further improved.