229 results about "Algorithms performance" patented technology

The invention relates to the technical field of integrated energy micro-grid scheduling optimization, and discloses an improved TD3 algorithm-based electricity-hydrogen-heat integrated energy micro-grid optimization scheduling method, which aims at minimizing the sum of the environmental cost, the equipment life loss cost and the large power grid interaction cost, and uses the improved TD3 algorithm to optimize the power-hydrogen-heat integrated energy micro-grid scheduling optimization. The optimization scheduling strategy of the electricity-hydrogen-heat comprehensive energy microgrid system is researched, a hydrogen energy storage mathematical model is constructed, the electrothermal coupling characteristic between the hydrogen energy storage system and the microgrid is explored, the optimization scheduling process of the microgrid is converted into a Markov decision process, and an experience playback mechanism is introduced to improve the algorithm performance. According to the method, the environmental friendliness and the scheduling strategy solving speed of the micro-grid energy storage system are improved, and the solving speed of the optimal scheduling strategy is accelerated by using the improved priority experience playback reinforcement learning algorithm.

The invention provides a charging station dynamic pricing method based on a game grey wolf algorithm, and aims to solve the problem of dynamic pricing of a single charging station. According to the method, modeling is carried out on a charging station dynamic pricing problem, a Markov decision process is combined, a double-layer gradual-reduction population reduction strategy is introduced, the search range is gradually reduced, and the global search efficiency of an algorithm is improved; meanwhile, in combination with a self-adaptive local search strategy, the local search strength is dynamically adjusted to solve the optimal solution as much as possible. In order to further improve the performance of the algorithm, the neural plasticity variation strategy is fused, the capability of the algorithm for jumping out of local optimum is enhanced, a competitive cooperation mechanism based on hierarchical game is designed, and the global search and local convergence capability among populations is balanced. The improvements effectively improve the convergence, stability and robustness of the algorithm. Experimental results show that the method has good solution precision and convergence efficiency under dynamic requirements, the income optimization capability of the charging station and the accuracy of dynamic pricing are remarkably improved, and an efficient and practical solution is provided for the dynamic pricing problem of the charging station.

The invention relates to the technical field of sparse radar depth completion, in particular to a sparse radar depth completion method based on Boolean mask constraint. The method comprises the following steps: generating a plurality of Boolean masks, processing the Boolean masks, and superposing the processed areas into a depth map; performing depth reversal operation to obtain a reversed depth value; performing multi-scale convolution kernel expansion operation to obtain an expanded depth map; performing mask-constrained hole closing operation to obtain a depth map after hole closing; performing depth value upward extension operation to obtain an extended depth map; performing mask-constrained hole filling operation to obtain a depth map after hole filling; performing median filtering and Gaussian filtering operation on the depth map after hole filling to obtain a filtered depth map; and performing depth map reversal operation on the filtered depth map to obtain a final depth map. The method can effectively protect the deep fracture area, and improves the performance of the depth completion algorithm.

The invention discloses a visual defect detection method for precise and complex parts, and particularly relates to the technical field of part detection. The method comprises the following steps: dividing the surface of a precise complex part into a plurality of to-be-detected areas, calculating a gray level co-occurrence matrix characteristic value, texture characteristic distribution and a spectrum analysis result by combining image data under multi-angle and multi-light-source conditions, further comparing surface bulges and gray level difference characteristics of a normal area and the to-be-detected areas, and calculating the surface defect similarity. On the basis of surface texture complexity and defect similarity, risk misjudgment is performed through fuzzy logic analysis, a detection area is divided into high, medium and low risk areas, a detection algorithm is optimized to improve accuracy, finally, algorithm performance is verified through test data, a parameter adjustment range is limited when the misjudgment rate is increased, systematic deviation and performance reduction are avoided, and the detection accuracy is improved. Therefore, the detection stability and the overall quality control reliability are improved.

The invention discloses an unmanned aerial vehicle three-dimensional path planning method and system based on a multi-strategy improved black-wing optimization algorithm, and the method comprises the steps: generating an initial population through a Latin hypercube sampling method, thereby improving the distribution uniformity and diversity of the population; an adaptive weight factor is introduced to realize dynamic balance of exploration and development capabilities; meanwhile, a dynamic reverse learning strategy is combined, so that the global search capability is effectively enhanced, and premature convergence is avoided; the algorithm performance is further improved by fusing a warning person position updating formula in a sparrow search algorithm. The method is used for solving the problems that a traditional path planning problem is prone to falling into local optimum, the convergence speed is low, and the path is unstable. And a shorter and safer optimal flight path can be efficiently planned.

The invention discloses a low ionosphere intelligent inversion method based on VLF radio wave detection, and the method comprises the steps: collecting ground-based very-low-frequency wave amplitude data, defining a gated cycle unit network objective function as an optimized performance index, and employing a mean square error as the objective function; and carrying out dynamic position updating by adopting a whale optimization algorithm, setting a double-stop criterion, and stopping iterative training to obtain an optimal model when the maximum number of iterations or a verification error convergence threshold value is reached. The calculation result of the LWPC model is used as a training set, the GRU network is used as an objective function estimator from foundation very low frequency data to low ionosphere parameters, the inversion method can accurately approach a real model, and the algorithm performance is effective and feasible and is remarkably improved compared with an existing model.

The invention discloses an automatic machine learning collaborative optimization method based on hardware perception. The method is used for design of an integrated learning field programmable gate array accelerator in edge artificial intelligence. The method aims at solving the balance challenge between algorithm performance and hardware resource consumption. The method comprises the following steps: defining a joint optimization space containing algorithm hyper-parameters and FPGA hardware configuration parameters; a collaborative optimization objective function is established, the function is determined by algorithm performance indexes and predicted FPGA hardware resource occupancy, and resource prediction quantifies hardware influence according to algorithm hyper-parameters; and searching according to the objective function in the joint optimization space by utilizing an AutoML optimization algorithm so as to identify the optimal configuration of the balance performance and resources. According to the method, the problems of low resource efficiency and high development cost caused by separate optimization of an algorithm and hardware in the prior art are effectively solved, balanced design is realized, automation is enhanced, and complexity is reduced.

The invention provides a robot vision object semantic understanding and posture generation method based on a large model, and belongs to the technical field of image processing. Comprising the following steps: S1, inputting an image and a 3D model; s2, object detection; s3, multi-modal feature alignment is carried out; s4, diffusion model sampling; s5, performing geometric screening; s6, carrying out NeRF (New Random Field) morphological modeling; s7, optimizing the optical flow; s8, joint loss calculation; s9, confidence coefficient analysis; and S10, outputting the attitude junction. According to the invention, through an innovative single-view rendering-optical flow optimization closed loop strategy, the calculation overhead is significantly reduced and the estimation precision is improved. According to the method, the algorithm performance in a complex scene is remarkably improved through multi-scale feature fusion and a confidence decomposition strategy.

The invention discloses an unmanned aerial vehicle cluster safety path planning method based on an improved black-wing optimization algorithm, and the method comprises the steps: introducing an anti-collision constraint for each unmanned aerial vehicle in an unmanned aerial vehicle cluster, and forming a multi-factor constraint flight total cost function of the unmanned aerial vehicle cluster; introducing a good point set theory, an optimal value guide differential mutation strategy and an elite learning mechanism to improve a black wing plinary optimization algorithm; taking a multi-factor constraint flight total cost function of the unmanned aerial vehicle cluster as a fitness function, and solving the fitness function by adopting an improved black wing plinth algorithm; and simulating the path planning of the unmanned aerial vehicle, and verifying the effectiveness of the proposed algorithm. According to the algorithm, better performance can be obtained in the face of complex problems; the algorithm has flexible adjustment capability, the path is updated according to the information obtained in real time, learning can be carried out by comparing the difference between the algorithm and elite individuals and the difference between the algorithm and the population average value in the path planning process, and the algorithm performance is improved.

The invention discloses a multi-dimensional multi-layer intelligent identification algorithm evaluation index system construction method, and belongs to the technical field of intelligent perception algorithm performance verification evaluation, and the method comprises the steps: constructing a traditional evaluation index based on the target identification precision, the target omission rate, the target false alarm rate and the time delay; constructing a generalization evaluation index based on the target adaptability index and the environmental adaptability index; constructing a safety evaluation index based on the confrontation sample safety index and the poisoning sample safety index; constructing a learning ability evaluation index based on the inversion learning ability index, the few-sample learning ability index and the incremental learning ability index; based on a traditional evaluation index, a generalization evaluation index, a safety evaluation index, a learning ability evaluation index and corresponding weights, a comprehensive evaluation result is obtained through calculation, so that the problem that a performance verification evaluation method of an intelligent sensing algorithm at the present stage is inaccurate, incomplete and unreliable in evaluation is solved.

The invention discloses a mechanical arm simulation system supporting industrial quality inspection, which relates to the technical field of industrial automation and comprises a multi-physics field simulation engine module, a digital twin interface module, a defect sample library module and a cooperative control module. According to the method, traditional entity mechanical arm debugging is replaced by rigid body dynamics simulation based on a physical engine, loss and production line shutdown caused by repeated trial and error of equipment are avoided, rapid construction of the digital twin environment of the production line is achieved by means of an efficient CAD model conversion process and a lightweight loading technology, and high-fidelity optical simulation is achieved through a parameterized defect model library and high-fidelity optical simulation. Various common industrial defects and complex optical effects are covered, the modeling bottleneck of a quality inspection scene is broken through, real-time closed-loop interaction of visual inspection and motion control is realized by using thread separation and an efficient communication technology, and the recognition robustness of a visual algorithm to tiny defects is guaranteed; the reliability of algorithm verification is improved through the quantitative output of algorithm performance indexes and the comprehensive test of extreme working conditions by the simulation system.

The present invention discloses a small-sample target detection method and system for SAR images based on meta-learning, belonging to the technical field of SAR image target detection, which includes the following steps: S1: Data collection and production; S2: Model rough tuning; S3: Model fine tuning; S4: Target detection. The present invention can realize the fusion of multi-scale feature information between the features of the SAR image itself and the corresponding scattering characteristic map, enabling the model to obtain more effective features; the feature fusion part of the small-sample target detection algorithm is optimized. In particular, when calculating the M2 feature, the hybrid attention representation M1 of the support set and the query set is used, which can more fully explore the feature connection between the support set and the query set, screen out the most effective features and reduce the weight of inefficient features, making the algorithm performance index higher.

The invention discloses an imitated radio frequency division multiplexing signal detection method based on low-complexity optimization expected propagation, and belongs to the signal detection technology of multi-carrier communication. According to the method, information is updated through iteration to approach an ML result, and the complexity is reduced from an exponential level to a cubic level; in order to reduce the complexity of the EP, sparse and quasi-banded channel structures in the DAFT domain are utilized. Through matrix block decomposition, upper and lower triangular decomposition and upper triangular matrix forward substitution, low-complexity matrix inversion is successfully realized, and the complexity of the EP algorithm is reduced to a linear level. In addition, the EP algorithm performance is further improved through a deep learning optimization moment matching strategy, so that the process becomes more adaptive through trainable positive and negative component parameters. In addition, DAFT domain iteration detection and decoding can be achieved based on the method, external information of a decoder is fed back to a detector, and the reliability of the whole system is remarkably improved. According to the invention, effective balance between performance improvement and calculation complexity is realized.

The invention belongs to the technical field of traditional Chinese medicine information processing, and provides a traditional Chinese medicine disease prescription data base construction method and system.The method comprises the steps that five types of data including ancient literatures, modern medical records, prescription data, traditional Chinese medicine resources and medical record manuscript images are accessed through an extensible adapter frame; a traditional Chinese medicine term service platform is constructed by using multi-source term resources, and standardized data is generated from the analyzed data; the method comprises the following steps: extracting diseases, syndromes, prescriptions and traditional Chinese medicine entities by adopting a rule engine and a domain self-adaptive entity recognition model, on the basis of a public traditional Chinese medicine standard knowledge graph, aggregating neighbor node information through GAT, and introducing a contrast learning optimization entity relationship classifier to obtain an enhanced entity vector and classify a semantic relationship; constructing a knowledge graph based on the extracted semantic relationship; and carrying out data quality and algorithm performance monitoring and tracing on data access analysis, standardization processing, knowledge extraction and graph construction. Uniform access and standardized integration of multi-source heterogeneous traditional Chinese medicine data are realized.

The invention discloses a deconvolution super beam forming method and system for a multi-beam sonar image, and belongs to the field of sonar image processing. Collecting multi-beam sonar data, and creating a target characteristic database; generating a point spread function for deconvolution by using a super beam forming method; the azimuth spectrum is optimized through a Richardson-Lucy iterative algorithm; adjusting algorithm parameters, and verifying the performance of the optimized Dcv-HBF algorithm through simulation and experiments; and outputting a clear target sonar image for target detection and background noise suppression analysis. According to the method, the principle of hyper-beam forming HBF and the principle of deconvolution conventional beam forming Dcv-CBF are combined, an HBF beam pattern serves as a point spread function PSF, deconvolution is conducted on the orientation spectrum of the HBF through the R-L iterative algorithm, the spatial spectrum of Dcv-HBF is generated, and the spatial resolution and background noise suppression capacity of sonar imaging are improved.

The invention discloses a medical image generation method based on a diffusion model, which introduces a novel method based on a potential diffusion model LDM, and uses a tumor mask and a CT image as condition input to generate a high-quality PET image containing a tumor focus. According to the method, an image is coded by using a vector quantization variational auto-encoder VQVAE, denoising is performed in a hidden space by using a diffusion model, and a generation process is guided by using a conditional encoder. The real tumor with controllable shape, size and position can be generated, and the structure and metabolic activity of the real tumor can be simulated. A comparison experiment with a real PET image shows that the generated image has high fidelity, and the potential of the method for enhancing the performance of the AI algorithm in a data limited environment is highlighted.

This invention relates to a Transformer-based gait emotion recognition method, apparatus, electronic device, and storage medium. The Transformer-based gait emotion recognition method of this invention includes: acquiring a gait video of an emotion to be identified; preprocessing the gait video to obtain a continuous gait sequence; inputting the gait sequence into a trained gait emotion recognition network to obtain the emotion category corresponding to the gait video; wherein the gait emotion recognition network is an autoencoder network model based on Spatial-Temporal Transformer. The Transformer-based gait emotion recognition method of this invention introduces and applies the Transformer algorithm, designing an autoencoder network model based on Spatial-Temporal Transformer, which significantly improves the algorithm performance.

The invention belongs to the technical field of digital image processing, and particularly relates to a self-adaptive layered confocal intracranial microwave imaging method and device and a storage medium. When antenna arrays with different antenna numbers and antenna arrangement structures are used, the algorithm can perform self-adaptive optimization antenna selection; the optimal calculation antenna pair is selected for imaging at different target positions, and intracranial layered imaging is achieved; the method comprises the following steps: calculating the transmission time when a microwave signal is transmitted from an antenna # imgabs0 # and received by an antenna # imgabs2 # after being reflected at a target point # imgabs1 #, calculating the propagation speed of the signal in the brain, introducing a gain coefficient, carrying out time shifting summation on reflected signals received by different antennas, reconstructing a relative grayscale image of a target area, and detecting an intracranial hemorrhage area. The imaging process is optimized by adaptively selecting the antenna combination, the algorithm performance and the image quality are improved, and the positioning error is reduced.

The invention relates to the technical field of passwords, in particular to an Ascon algorithm optimization implementation method based on an RISC-V processor, and the method comprises the steps: reconstructing S-box substitution logic through RISC-V assembly; 64-bit cyclic shift is optimized through a bit interleaving technology; fully expanding assembly codes to eliminate cycle overhead; the intermediate state and the key are stored in a register. According to the method, the execution efficiency and performance of the Ascon algorithm on an RISC-V platform are remarkably improved through multiple technical means, redundant bit operation is effectively reduced through reconstruction of S-box substitution logic, calculation delay is reduced through optimization of bit operation sequences and multiplexing of temporary variables, 64-bit cyclic shift operation is decomposed through the bit interleaving technology, and the calculation efficiency of the algorithm is improved. The operation is more efficient, the overhead of loop control is eliminated through an assembly-level full-expansion technology, unnecessary jump instructions are avoided, the execution speed is further improved, and the problems of insufficient Ascon algorithm performance and resource waste caused by low utilization rate of the RISC-V architecture are effectively solved.

The invention belongs to the technical field of algorithm component upgrading, and relates to an algorithm component remote upgrading method, which is characterized in that a data acquisition module established based on an SCADA or EMS system monitors and collects performance data of each algorithm component in real time, and the system can adapt to the change of a power grid environment in real time by setting and dynamically adjusting a performance threshold value. The algorithm performance is ensured to be always in a reasonable range, if the performance of a certain algorithm component is lower than a threshold value, the system automatically triggers a data collection process, and an anomaly detection algorithm is applied to identify data deviation based on collected input data, output results and operation log information, so that a new data set is constructed through comprehensive analysis; the corresponding optimization strategy is generated through the automatic optimization algorithm based on the data set, the algorithm component upgrade package is created through the generated optimization strategy, and remote upgrade operation is performed, so that the problems of insufficient manual intervention, slow response, inaccurate optimization and the like in the prior art are solved, and the intelligent level and the adaptive capacity of the power grid monitoring system are remarkably improved.

The invention discloses an efficient routing algorithm for offshore buoy networking. The problem that existing offshore buoy networking information transmission is poor in real-time performance and stability is mainly solved. According to the scheme, the method comprises the following steps: inputting the position of an offshore buoy network, constructing a communication link connectivity model, and abstracting a network link topology by means of a graph theory representation mode; comprehensively considering network efficiencies such as time delay, packet loss rate and network load balancing, and determining link cost between two buoy nodes; aiming at the defect that the ant colony algorithm is easy to fall into local optimum, two-point improvement is carried out on the basic ant colony algorithm. The method comprises the following steps: firstly, during construction of initial pheromones, in a first iteration process of an algorithm, 20% of ants in a population are selected, a path with the minimum cost is selected based on a neighbor algorithm, and the remaining ants adopt random selection paths, so that the convergence of the algorithm is improved, and meanwhile, the diversity of the population is ensured; and a second point, aiming at the problem that a large number of ants are easy to walk to a repeated path when the ant colony is converged to the local optimum in the later stage of iteration, adding repeated check operation in an algorithm updating process. According to the specific operation, when 20% of ants select repeated paths, the ants of the repeated paths select new paths in a random mode, the path search space is expanded, and the algorithm performance is improved. According to the method, the efficient route of buoy networking communication can be dynamically planned, and the real-time performance and the stability of the network are met.

The invention provides a multi-sound-source arrival direction estimation method and device based on a frequency focusing spatial spectrum. The method comprises the following steps: firstly, generating microphone array data and constructing a training data set by using a preset simulation environment and a microphone array signal simulator; secondly, constructing a mask estimation network model through sound source masking training; then, sub-bandwidth division is carried out on the enhanced multi-sound-source signals in a frequency band splicing mode, and a weighted focusing covariance matrix is obtained; thirdly, broadband focusing spatial spectrum features are constructed through a sub-bandwidth frequency focusing covariance matrix obtained through beam forming, and a convolutional neural network is constructed based on the features; and finally, according to the mask estimation network model and the multi-sound-source DOA spatial spectrum estimation network model, decoding to obtain a multi-sound-source DOA spatial spectrum, and completing multi-sound-source DOA estimation. According to the invention, the intermediate representation based on the broadband frequency focusing spatial spectrum reduces the influence of nonlinear distortion of the neural network on DOA estimation; and the stability of algorithm performance and the fault tolerance of DOA estimation are improved.

The invention belongs to the technical field of multi-source information fusion navigation, and particularly relates to a multi-source information fusion navigation method based on a variational Bayesian factor graph, a factor graph model is constructed by combining factor nodes of an inertial navigation system (INS), a scene matching navigation system (IMNS) and a global navigation satellite system (GNSS), and navigation state information is obtained according to a factor graph reasoning method. In order to solve the problem that the performance of a traditional multi-source information fusion navigation algorithm is reduced due to the fact that noise is unknown or suddenly changed, observation noise is modeled into Student-T distribution, then the system state is predicted, and the navigation state and unknown noise parameters are iterated by means of a Gaussian Newton iteration method and variational Bayesian learning. According to the invention, the algorithm is high in universality, has the characteristics of plug and play, and is high in expansibility and robustness.

The invention provides an artificial cochlea objective debugging method and system based on wide-spectrum natural sound evoked electroencephalography, and the method comprises the steps: collecting an auditory electroencephalogram of a tested person under the stimulation of wide-spectrum natural scale sound synthesized by a plurality of vocoders, and carrying out the preprocessing of the auditory electroencephalogram, and obtaining an auditory electroencephalogram data set; training the deep learning network model by using data in the auditory electroencephalogram data set to obtain a trained deep learning network model; the deep learning network model is provided with two convolution layers for extracting features including time frequency and channel information of an auditory electroencephalogram, and is provided with a full connection layer for classification to obtain scale sound sensing electroencephalogram decoding and prediction accuracy; and quantizing different vocoder algorithm performances according to the obtained scale sound sensing electroencephalogram decoding and prediction accuracy, and objectively debugging the artificial cochlea based on the quantized different vocoder algorithm performances.

The invention discloses a YOLO deep learning-based medial temporal lobe atrophy grade automatic scoring system. The system comprises a coronal section screening module, an improved YOLOv8 model construction module, an optimal coronal section automatic selection module and a medial temporal lobe atrophy detection and automatic scoring module. The invention provides an automatic MTA scoring technology which gives consideration to accuracy, universality and operation convenience. By using a lightweight deep learning YOLO target detection framework and combining with a proper loss function, the calculation overhead is reduced while the algorithm performance is ensured, and by combining with a friendly front-end interaction interface, the efficiency and accuracy of clinical MTA scoring are improved. The automatic scoring system for the medial temporal lobe atrophy is efficient, accurate and consistent, has remarkable clinical application value and popularization prospects, and is expected to play an important role in early diagnosis and monitoring of neurodegenerative diseases such as Alzheimer's disease.

The invention discloses a vertical mill hydraulic control method and system based on an improved dung beetle optimization algorithm, and the method comprises the steps: constructing a vertical mill machine-liquid coupling simulation model, introducing a fusion good point set and reverse learning strategy, a self-adaptive breeding theft strategy and a self-adaptive hybrid variation strategy into the dung beetle algorithm, and obtaining a simulation model; the quantization factor and the scaling factor of the fuzzy PID controller are optimized through IDBO and are applied to the hydraulic control system of the vertical mill. Control parameters can be dynamically adjusted to adapt to different working conditions and disturbances; the longitudinal fluctuation error of the grinding roller is effectively reduced; the good point set and the reverse learning strategy are fused to initialize the population, the quality and diversity of the initial population solution are improved, the premature convergence phenomenon is effectively avoided, the blank area in the search space is reduced, and the algorithm performance is improved; meanwhile, an adaptive reproduction theft strategy is introduced, the diversity of the search range is increased, the global search ability of the algorithm is improved in the earlier stage of iteration, and the algorithm is endowed with stronger local fine search ability in the later stage of iteration.

The invention relates to an LDPC (Low Density Parity Check) quantitative hierarchical decoding working method combining an improved dragonfly algorithm with residual degradation, and belongs to the technical field of wireless communication. The method specifically comprises the steps that firstly, a pilot signal is sent by a signal source, average least square estimation of channel state information is obtained through analysis based on representation of a received signal in a time domain and a frequency domain, noise power is determined, then the pilot signal is introduced into an improved dragonfly algorithm, quantization parameter set training is carried out, and the improved dragonfly algorithm reads a sample signal to obtain the channel state information. And finally, applying the self-adaptive quantization parameter set obtained by training to a residual degradation decoding algorithm, and if a decoding result does not meet a verification equation, carrying out next iteration. The performance of the residual degradation decoding algorithm is superior to that of the DLNMS, hardware resources are saved, the decoding performance of layered quantization is improved, the decoding time is shortened, and therefore the decoding speed is increased.

The invention relates to an LDPC (Low Density Parity Check) decoder oriented to a CCSDS (Consultative Committee for Space Data System) standard and a decoding method thereof, relates to the technical field of satellite communication and channel coding, and aims to solve the technical problems of insufficient parallelism and low resource utilization rate of a satellite communication LDPC decoder in the prior art. The LDPC decoder for the CCSDS standard comprises a frame input module, a control module, an initial probability storage module, a check node calculation module, a variable node calculation module and a frame output module. The LDPC decoder facing the CCSDS standard and the decoding method thereof are suitable for efficient decoding of long-frame data in satellite communication, dynamic block storage architecture, hybrid iterative decoding algorithm and configurable control module collaborative design, balance of decoding algorithm performance and complexity is achieved through storage design optimization, the algorithm flexibility is improved through a configurable strategy, and the decoding efficiency is improved. And the contradiction between the efficiency and resources of long-frame LDPC decoding in satellite communication is systematically solved.

The invention relates to a two-dimensional code image classification method based on double-domain contrast learning and sample adaptive enhancement, and belongs to the field of semi-supervised image classification. According to the method, a double-domain contrast learning module (Freq-MOCO) is designed. The module carries out comparative learning in parallel in a feature domain and a frequency domain, the feature domain learning mainly focuses on spatial features of images, and the frequency domain learning focuses on spectrum features and global structure information. Besides, a sample adaptive enhancement module (SAA) is also designed, and the learning effect of the samples is further improved by identifying the simple samples and performing more diversified enhancement on the simple samples. The SAA module selects a simple sample by using historical loss information, and applies a richer enhancement strategy to the simple sample, so as to ensure that the sample which cannot effectively promote model learning originally can generate positive influence on model training through diversified enhancement modes. The algorithm provided by the invention has better performance than several recent algorithms.

The invention discloses a mobile terminal lightweight drug taking auxiliary detection method and system based on a face image, and the method comprises the following steps: S1, collecting a face image of a detected person through a camera, and carrying out the preprocessing operation of face detection, background removal and face alignment; s2, designing a lightweight neural network module, and constructing a lightweight drug taking detection network based on the module; s3, acquiring a face data set of a drug addict and training a drug addict detection network; s4, extracting face features by using the drug taking detection network and obtaining a classification result and a certainty degree; and S5, converting and transplanting the drug taking detection network model to a mobile terminal, and constructing a mobile terminal application. According to the method, the drug addicts are detected based on the facial images, the defects that a traditional biochemical sample detection method is high in matching requirement, long in detection period, difficult to screen in a large range and the like are overcome, meanwhile, deep learning and lightweight thoughts are used, the algorithm performance is effectively improved, and the method is suitable for mobile terminal application scenes, portable application scenes and other application scenes.