Spatiotemporal Information Processing Research Articles

Two-Terminal Neuromorphic Devices for Spiking Neural Networks: Neurons, Synapses, and Array Integration.

The ever-increasing volume of complex data poses significant challenges to conventional sequential global processing methods, highlighting their inherent limitations. This computational burden has catalyzed interest in neuromorphic computing, particularly within artificial neural networks (ANNs). In pursuit of advancing neuromorphic hardware, researchers are focusing on developing computation strategies and constructing high-density crossbar arrays utilizing history-dependent, multistate nonvolatile memories tailored for multiply-accumulate (MAC) operations. However, the real-time collection and processing of massive, dynamic data sets require an innovative computational paradigm akin to that of the human brain. Spiking neural networks (SNNs), representing the third generation of ANNs, are emerging as a promising solution for real-time spatiotemporal information processing due to their event-based spatiotemporal capabilities. The ideal hardware supporting SNN operations comprises artificial neurons, artificial synapses, and their integrated arrays. Currently, the structural complexity of SNNs and spike-based methodologies requires hardware components with biomimetic behaviors that are distinct from those of conventional memristors used in deep neural networks. These distinctive characteristics required for neuron and synapses devices pose significant challenges. Developing effective building blocks for SNNs, therefore, necessitates leveraging the intrinsic properties of the materials constituting each unit and overcoming the integration barriers. This review focuses on the progress toward memristor-based spiking neural network neuromorphic hardware, emphasizing the role of individual components such as memristor-based neurons, synapses, and array integration along with relevant biological insights. We aim to provide valuable perspectives to researchers working on the next generation of brain-like computing systems based on these foundational elements.

A Spatial-Channel-Temporal-Fused Attention for Spiking Neural Networks.

Spiking neural networks (SNNs) mimic brain computational strategies, and exhibit substantial capabilities in spatiotemporal information processing. As an essential factor for human perception, visual attention refers to the dynamic process for selecting salient regions in biological vision systems. Although visual attention mechanisms have achieved great success in computer vision applications, they are rarely introduced into SNNs. Inspired by experimental observations on predictive attentional remapping, we propose a new spatial-channel-temporal-fused attention (SCTFA) module that can guide SNNs to efficiently capture underlying target regions by utilizing accumulated historical spatial-channel information in the present study. Through a systematic evaluation on three event stream datasets (DVS Gesture, SL-Animals-DVS, and MNIST-DVS), we demonstrate that the SNN with the SCTFA module (SCTFA-SNN) not only significantly outperforms the baseline SNN (BL-SNN) and two other SNN models with degenerated attention modules, but also achieves competitive accuracy with the existing state-of-the-art (SOTA) methods. Additionally, our detailed analysis shows that the proposed SCTFA-SNN model has strong robustness to noise and outstanding stability when faced with incomplete data, while maintaining acceptable complexity and efficiency. Overall, these findings indicate that incorporating appropriate cognitive mechanisms of the brain may provide a promising approach to elevate the capabilities of SNNs.

Dimensionality reduction in bulk-boundary reaction-diffusion systems.

Intracellular protein patterns regulate many vital cellular functions, such as the processing of spatiotemporal information or the control of shape deformations. To do so, pattern-forming systems can be sensitive to the cell geometry by means of coupling the protein dynamics on the cell membrane to dynamics in the cytosol. Recent studies demonstrated that modeling the cytosolic dynamics in terms of an averaged protein pool disregards possibly crucial aspects of the pattern formation, most importantly concentration gradients normal to the membrane. At the same time, the coupling of two domains (surface and volume) with different dimensions renders many standard tools for the numerical analysis of self-organizing systems inefficient. Here, we present a generic framework for projecting the cytosolic dynamics onto the lower-dimensional surface that respects the influence of cytosolic concentration gradients in static and evolving geometries. This method uses a priori physical information about the system to approximate the cytosolic dynamics by a small number of dominant characteristic concentration profiles (basis), akin to basis transformations of finite element methods. As a proof of concept, we apply our framework to a toy model for volume-dependent interrupted coarsening, evaluate the accuracy of the results for various basis choices, and discuss the optimal basis choice for biologically relevant systems. Our analysis presents an efficient yet accurate method for analyzing pattern formation with surface-volume coupling in evolving geometries.

A Single-Pixel Event Photoactive Device for Real-Time, In-Sensor Spatiotemporal Optical Information Processing.

The increasing demand for energy-efficient, sophisticated optical sensing technologies in various applications, from machine vision to optical communication, highlights the necessity for innovations in spatiotemporal information sensing and processing at a nearly single-pixel scale. Traditional methods, including multi-pixel photodetector arrays and event-based camera systems, often fail to provide rapid, real-time detection and processing of dynamic events within the sensor. This shortcoming is particularly notable in handling high-dimensional spatiotemporal data, where the dependency on sequential data input and external processing tools leads to latency, reduced throughput, and heightened energy consumption, thereby impeding real-time parallel data processing capabilities. Here, a carrier-selective, single-pixel, position-sensitive planar photoactive device that integrates spatiotemporal event sensing with inherent short-term memory capabilities is introduced. The proof-of-concept single-pixel event photoactive device enables in-sensor spatiotemporal parallel optical information processing, efficiently managing multibit (>4 bit) data simultaneously and facilitating ultrafast (≈0.4µs) recognition of input patterns with low energy consumption (25 femtojoules). Additionally, by adjusting the operating speed from continuous to pulsed light illumination, the sensor array can detect trajectories and absolute position of events, offering in-sensor optical flow detection. This single-pixel event photodetector marks significant advancement toward developing compact, energy-efficient, ultrafast sensors suitable for a wide range of in sensor-based photonic applications.

STCSNN: High energy efficiency spike-train level spiking neural networks with spatio-temporal conversion

Brain-inspired spiking neuron networks (SNNs) have attracted widespread research interest due to their low power features, high biological plausibility, and strong spatiotemporal information processing capability. Although adopting a surrogate gradient (SG) makes the non-differentiability SNN trainable, achieving comparable accuracy for ANNs and keeping low-power features simultaneously is still tricky. In this paper, we proposed an energy-efficient spike-train level spiking neural network with spatio-temporal conversion, which has low computational cost and high accuracy. In the STCSNN, spatio-temporal conversion blocks (STCBs) are proposed to keep the low power features of SNNs and improve accuracy. However, STCSNN cannot adopt backpropagation algorithms directly due to the non-differentiability nature of spike trains. We proposed a suitable learning rule for STCSNNs by deducing the equivalent gradient of STCB. We evaluate the proposed STCSNN on static and neuromorphic datasets, including Fashion-Mnist, Cifar10, Cifar100, TinyImageNet, and DVS-Cifar10. The experiment results show that our proposed STCSNN outperforms the state-of-the-art accuracy on nearly all datasets, using fewer time steps and being highly energy-efficient.

Oxide Ionic Neuro-Transistors for Bio-inspired Computing.

Current computing systems rely on Boolean logic and von Neumann architecture, where computing cells are based on high-speed electron-conducting complementary metal-oxide-semiconductor (CMOS) transistors. In contrast, ions play an essential role in biological neural computing. Compared with CMOS units, the synapse/neuron computing speed is much lower, but the human brain performs much better in many tasks such as pattern recognition and decision-making. Recently, ionic dynamics in oxide electrolyte-gated transistors have attracted increasing attention in the field of neuromorphic computing, which is more similar to the computing modality in the biological brain. In this review article, we start with the introduction of some ionic processes in biological brain computing. Then, electrolyte-gated ionic transistors, especially oxide ionic transistors, are briefly introduced. Later, we review the state-of-the-art progress in oxide electrolyte-gated transistors for ionic neuromorphic computing including dynamic synaptic plasticity emulation, spatiotemporal information processing, and artificial sensory neuron function implementation. Finally, we will address the current challenges and offer recommendations along with potential research directions.

ALBSNN: ultra-low latency adaptive local binary spiking neural network with accuracy loss estimator.

Spiking neural network (SNN) is a brain-inspired model with more spatio-temporal information processing capacity and computational energy efficiency. However, with the increasing depth of SNNs, the memory problem caused by the weights of SNNs has gradually attracted attention. In this study, we propose an ultra-low latency adaptive local binary spiking neural network (ALBSNN) with accuracy loss estimators, which dynamically selects the network layers to be binarized to ensure a balance between quantization degree and classification accuracy by evaluating the error caused by the binarized weights during the network learning process. At the same time, to accelerate the training speed of the network, the global average pooling (GAP) layer is introduced to replace the fully connected layers by combining convolution and pooling. Finally, to further reduce the error caused by the binary weight, we propose binary weight optimization (BWO), which updates the overall weight by directly adjusting the binary weight. This method further reduces the loss of the network that reaches the training bottleneck. The combination of the above methods balances the network's quantization and recognition ability, enabling the network to maintain the recognition capability equivalent to the full precision network and reduce the storage space by more than 20%. So, SNNs can use a small number of time steps to obtain better recognition accuracy. In the extreme case of using only a one-time step, we still can achieve 93.39, 92.12, and 69.55% testing accuracy on three traditional static datasets, Fashion- MNIST, CIFAR-10, and CIFAR-100, respectively. At the same time, we evaluate our method on neuromorphic N-MNIST, CIFAR10-DVS, and IBM DVS128 Gesture datasets and achieve advanced accuracy in SNN with binary weights. Our network has greater advantages in terms of storage resources and training time.

A novel brain-inspired hierarchical perception-association circuit based on memristor arrays

In the brain, primary sensory cells can efficiently perceive multimodal stimuli, and then associative memory cells perform an advanced bidirectional associative memory function with perceived information. Here, a brain-inspired hierarchical perception-association circuit based on memristor arrays is proposed. Firstly, a memristive reservoir computing circuit is designed to simulate a low-level perceptual function, which mainly comprises dynamic analog reservoirs, memristor arrays, and analog integrators, enabling efficient spatio-temporal information processing. Secondly, a spiking bidirectional associative memory memristive circuit, as the core of the whole circuit, is proposed to mimic a high-level associative function, which mainly consists of memristor arrays, current amplifiers, and leaky integrate-and-fire neuron circuits, implementing multimodal associative learning. The simulation results in LTspice show that the modified neuron circuit exhibits 66× improvement in energy efficiency, and the proposed circuit can precisely perform the letter association and vision–audio association tasks in a spiking fashion with an average firing energy consumption of 230 pJ/spike, which has a great potential to be embedded in mobile robotic platforms.

Ultralow‐Power Compact Artificial Synapse Based on a Ferroelectric Fin Field‐Effect Transistor for Spatiotemporal Information Processing

Artificial synapses are key elements in building bioinspired, neuromorphic computing systems. Ferroelectric field‐effect transistors (FeFETs) with excellent controllability and complementary metal oxide semiconductor (CMOS) compatibility are favorable to achieving synaptic functions with low power consumption and high scalability. However, because of the only nonvolatile ferroelectric (Fe) characteristics in the FeFET, it is difficult to develop bioplausible short‐term synaptic elements for spatiotemporal information processing. By judiciously combining defects (DE) and Fe domains in gate stacks, a compact artificial synapse featuring spatiotemporal information processing on a single Fe–DE fin FET (FinFET) is proposed. The devices are designed to work in a separate DE mode to induce short‐term plasticity by spontaneous charge detrapping, and a hybrid Fe–DE mode to trigger long‐term plasticity through the coupling of defects and Fe domains. The capability of the compact synapse is demonstrated by differentiating 16 temporal inputs. Moreover, the highly controllable static electricity of advanced FinFETs leads to an ultralow power of 2 fJ spike−1. An all Fe–DE FinFET reservoir computing (RC) system is then constructed that achieves a recognition accuracy of 97.53% in digit classification. This work enables constructing RC systems with fully advanced CMOS‐compatible devices featuring highly energy‐efficient and low‐hardware systems.

Traffic Flow Video Image Recognition and Analysis Based on Multi-Target Tracking Algorithm and Deep Learning

Traffic flow parameters are an important data support for the research and development of several technologies in the intelligent transportation system. Therefore, accurate and real-time estimation of traffic flow is particularly important for urban traffic. In this study, a real-time traffic flow detection system framework was constructed based on video image collection and analysis. According to the vehicle detection and tracking results, a traffic flow parameter estimation model and an improved LSTM network are proposed for spatiotemporal counting feature recognition. The results conclude that the developed framework can estimate the traffic flow density and count vehicles, as well as estimate the traffic flow velocity and traffic volume to estimate and optimize traffic flow, respectively. Additionally, the simulation results show that the proposed method can not only counts the two-way traffic vehicles quickly and accurately, but also avoids the use of the complex multi-target tracking method to spatiotemporal correlation of a single target, increases the speed and accuracy of the spatiotemporal information processing procedure, and has stronger scene adaptability.

Adaptively Lightweight Spatiotemporal Information-Extraction-Operator-Based DL Method for Aero-Engine RUL Prediction.

Accurate prediction of machine RUL plays a crucial role in reducing human casualties and economic losses, which is of significance. The ability to handle spatiotemporal information contributes to improving the prediction performance of machine RUL. However, most existing models for spatiotemporal information processing are not only complex in structure but also lack adaptive feature extraction capabilities. Therefore, a lightweight operator with adaptive spatiotemporal information extraction ability named Involution GRU (Inv-GRU) is proposed for aero-engine RUL prediction. Involution, the adaptive feature extraction operator, is replaced by the information connection in the gated recurrent unit to achieve adaptively spatiotemporal information extraction and reduce the parameters. Thus, Inv-GRU can well extract the degradation information of the aero-engine. Then, for the RUL prediction task, the Inv-GRU-based deep learning (DL) framework is firstly constructed, where features extracted by Inv-GRU and several human-made features are separately processed to generate health indicators (HIs) from multi-raw data of aero-engines. Finally, fully connected layers are adopted to reduce the dimension and regress RUL based on the generated HIs. By applying the Inv-GRU-based DL framework to the Commercial Modular Aero Propulsion System Simulation (C-MAPSS) datasets, successful predictions of aero-engines RUL have been achieved. Quantitative comparative experiments have demonstrated the advantage of the proposed method over other approaches in terms of both RUL prediction accuracy and computational burden.

Ultra-low latency spiking neural networks with spatio-temporal compression and synaptic convolutional block

Spiking neural networks (SNNs), as one of the brain-inspired models, has spatio-temporal information processing capability, low power feature, and high biological plausibility. The effective spatio-temporal feature makes it suitable for event streams classification. However, neuromorphic datasets, such as N-MNIST, CIFAR10-DVS, DVS128-gesture, need to aggregate individual events into frames with a new higher temporal resolution for event stream classification, which causes high training and inference latency. In this work, we proposed a spatio-temporal compression method to aggregate individual events into a few time steps of synaptic current to reduce the training and inference latency. To keep the accuracy of SNNs under high compression ratios, we also proposed a synaptic convolutional block to balance the dramatic changes between adjacent time steps. And multi-threshold Leaky Integrate-and-Fire (LIF) models with learnable membrane time constants are introduced to increase their information processing capability. We evaluate the proposed method for event stream classification tasks on neuromorphic N-MNIST, CIFAR10-DVS, and DVS128 gesture datasets. The experiment results show that our proposed method outperforms the state-of-the-art accuracy on nearly all datasets, using fewer time steps.

Multiscale feature fusion method for lane line detection based on time series

In order to solve lane line detection in difficult traffic conditions, such as shadow occlusion, signpost degradation, curves, and tunnels, numerous models have been proposed. However, most of the existing models conduct independent single-frame image detection, which makes it difficult to utilize the continuity of driving images and is ineffective in challenging scenes. To this end, we suggest a spatiotemporal information processing model for lane line recognition that enhances critical features. In order to properly learn the correlation between continuous images, we first employ a convolutional gated recurrent unit to process spatiotemporal driving information on the basis of U-Net. Second, the pyramid split attention (PSA) module is used to enhance or suppress the obtained feature expressions. Finally, the skip connection is used to fuse the features of different scales encoded by each stage with the features processed by PSA and gradually restore to the original image size. Experiments on the TuSimple dataset demonstrate that our model outperforms representative lane line detection networks in challenging driving scenes, with an F1-measure of up to 94.302%.

In materia implementation strategies of physical reservoir computing with memristive nanonetworks

Physical reservoir computing (RC) represents a computational framework that exploits information-processing capabilities of programmable matter, allowing the realization of energy-efficient neuromorphic hardware with fast learning and low training cost. Despite self-organized memristive networks have been demonstrated as physical reservoir able to extract relevant features from spatiotemporal input signals, multiterminal nanonetworks open the possibility for novel strategies of computing implementation. In this work, we report on implementation strategies of in materia RC with self-assembled memristive networks. Besides showing the spatiotemporal information processing capabilities of self-organized nanowire networks, we show through simulations that the emergent collective dynamics allows unconventional implementations of RC where the same electrodes can be used as both reservoir inputs and outputs. By comparing different implementation strategies on a digit recognition task, simulations show that the unconventional implementation allows a reduction of the hardware complexity without limiting computing capabilities, thus providing new insights for taking full advantage of in materia computing toward a rational design of neuromorphic systems.

Supervised Learning Algorithm Based on Spike Train Inner Product for Deep Spiking Neural Networks.

By mimicking the hierarchical structure of human brain, deep spiking neural networks (DSNNs) can extract features from a lower level to a higher level gradually, and improve the performance for the processing of spatio-temporal information. Due to the complex hierarchical structure and implicit nonlinear mechanism, the formulation of spike train level supervised learning methods for DSNNs remains an important problem in this research area. Based on the definition of kernel function and spike trains inner product (STIP) as well as the idea of error backpropagation (BP), this paper firstly proposes a deep supervised learning algorithm for DSNNs named BP-STIP. Furthermore, in order to alleviate the intrinsic weight transport problem of the BP mechanism, feedback alignment (FA) and broadcast alignment (BA) mechanisms are utilized to optimize the error feedback mode of BP-STIP, and two deep supervised learning algorithms named FA-STIP and BA-STIP are also proposed. In the experiments, the effectiveness of the proposed three DSNN algorithms is verified on the MNIST digital image benchmark dataset, and the influence of different kernel functions on the learning performance of DSNNs with different network scales is analyzed. Experimental results show that the FA-STIP and BP-STIP algorithms can achieve 94.73% and 95.65% classification accuracy, which apparently possess better learning performance and stability compared with the benchmark algorithm BP-STIP.

Depthwise Temporal Non-Local Network for Faster and Better Dynamic Hand Gesture Authentication

Dynamic hand gesture is an emerging and promising biometric trait. It contains both physiological and behavioral characteristics, which on the one hand can theoretically make authentication systems more accurate and more secure, and on the other hand can increase the difficulty of model design because it is essentially a fine-grained video understanding task. For authentication systems, equal error rate (EER) and real-time performance are two vital metrics. Current video understanding-based hand gesture authentication methods mainly focus on lowering the EER while neglecting to reduce the computational cost. In this paper, we propose a 2D CNN-based depthwise temporal non-local network (DwTNL-Net) that can take into account both EER and running efficiency. To enable the DwTNL-Net with spatiotemporal information processing capability, we design a temporal sharpening (TS) module and a DwTNL module for short- and long-term identity feature modeling, respectively. The TS module can assist the backbone in local behavioral characteristic understanding and can simultaneously remove redundant information and highlight behavioral cues while retaining sufficient physiological characteristics. In contrast, the DwTNL module focuses on summarizing global information and discovering stable patterns, which are finally used for local information enhancement. The complementary combination of our TS and DwTNL modules makes DwTNL-Net achieve substantial performance improvements. Extensive experiments on the SCUT-DHGA dataset and sufficient statistical analyses fully demonstrate the superiority and efficiency of our DwTNL-Net. The code is available at https://github.com/SCUT-BIP-Lab/DwTNL-Net.

Blended Glial Cell’s Spiking Neural Network

Spiking Neural Networks (SNNs), the third generation of artificial neural networks, have been widely employed. However, the realization of advanced artificial intelligence is challenging due to the dearth of efficient spatiotemporal information integration models. Inspired by brain neuroscientists, this paper proposes a novel spiking neural network - Blended Glial Cell’s Spiking Neural Network (BGSNN). BGSNN introduces glial cells as spatiotemporal information processing units based on neurons and synapses, and also provides four new network dynamics connection models which extend the information processing dimension, enhance the network global information integration in the spatiotemporal domain, as well as the plasticity of neurons and synapses. In this paper, a BGSNN application - Sudoku solver is designed and implemented on the "WenTian" neuromorphic prototype. On the Easybrain dataset, the BGSNN solver achieves 100% accuracy, outperforming the same structure SNN solver by 97% at the Evil difficulty level, and has faster converges speed compared with the SOTA Sudoku solver LSGA. On the kaggle dataset, the BGSNN solver achieves over 99.99% accuracy, outperforming the publicly available optimal DNN solver under this dataset by 3.82%. In addition, BGSNN exhibits good parallelism and sparsity, decreasing computation by at least 92.9% compared to serial solvers and reducing sparsity by 88% compared to the equal fully dense DNN. BGSNN improves the expression, feedback, and regulation capabilities of neural networks while maintaining the advantages of SNN parallel sparsity, making it simpler to implement advanced artificial intelligence.

Flexible multiterminal photoelectronic neurotransistors based on self‐assembled rubber semiconductors for spatiotemporal information processing

AbstractA significant step toward constructing high‐efficiency neuromorphic systems is the electronic emulation of advanced synaptic functions of the human brain. While previous studies have focused on mimicking the basic functions of synapses using single‐gate transistors, multigate transistors offer an opportunity to simulate more complex and advanced memory‐forming behaviors in biological synapses. In this study, a simple and general method is used to assemble rubber semiconductors into suspended two‐phase composite films that are transferred to the surface of the ion‐conducting membrane to fabricate flexible multiterminal photoelectronic neurotransistors. The suspended ion conductive film is used as the gate dielectrics and supporting substrate. The prepared devices exhibit excellent electrical stability and mechanical flexibility after being bent. Basic photoelectronic synaptic behavior and pulse‐dependent plasticity are emulated. Furthermore, the device realizes the spatiotemporally integrated electrical and optical stimuli to mimic spatiotemporal information processing. This study provides a promising direction for constructing more complex spiking neural networks and more powerful neuromorphic systems with brain‐like dynamic spatiotemporal processing functions.

A cerebellum inspired spiking neural network as a multi-model for pattern classification and robotic trajectory prediction.

Spiking neural networks were introduced to understand spatiotemporal information processing in neurons and have found their application in pattern encoding, data discrimination, and classification. Bioinspired network architectures are considered for event-driven tasks, and scientists have looked at different theories based on the architecture and functioning. Motor tasks, for example, have networks inspired by cerebellar architecture where the granular layer recodes sparse representations of the mossy fiber (MF) inputs and has more roles in motor learning. Using abstractions from cerebellar connections and learning rules of deep learning network (DLN), patterns were discriminated within datasets, and the same algorithm was used for trajectory optimization. In the current work, a cerebellum-inspired spiking neural network with dynamics of cerebellar neurons and learning mechanisms attributed to the granular layer, Purkinje cell (PC) layer, and cerebellar nuclei interconnected by excitatory and inhibitory synapses was implemented. The model's pattern discrimination capability was tested for two tasks on standard machine learning (ML) datasets and on following a trajectory of a low-cost sensor-free robotic articulator. Tuned for supervised learning, the pattern classification capability of the cerebellum-inspired network algorithm has produced more generalized models than data-specific precision models on smaller training datasets. The model showed an accuracy of 72%, which was comparable to standard ML algorithms, such as MLP (78%), Dl4jMlpClassifier (64%), RBFNetwork (71.4%), and libSVM-linear (85.7%). The cerebellar model increased the network's capability and decreased storage, augmenting faster computations. Additionally, the network model could also implicitly reconstruct the trajectory of a 6-degree of freedom (DOF) robotic arm with a low error rate by reconstructing the kinematic parameters. The variability between the actual and predicted trajectory points was noted to be ± 3 cm (while moving to a position in a cuboid space of 25 × 30 × 40 cm). Although a few known learning rules were implemented among known types of plasticity in the cerebellum, the network model showed a generalized processing capability for a range of signals, modulating the data through the interconnected neural populations. In addition to potential use on sensor-free or feed-forward based controllers for robotic arms and as a generalized pattern classification algorithm, this model adds implications to motor learning theory.

Short-Term Regional Temperature Prediction Based on Deep Spatial and Temporal Networks

Accurate prediction of air temperature is of great significance to outdoor activities and daily life. However, it is important and more challenging to predict air temperature in complex terrain areas because of prevailing mountain and valley winds and variable wind directions. The main innovation of this paper is to propose a regional temperature prediction method based on deep spatiotemporal networks, designing a spatiotemporal information processing module to align temperature data with regional grid points and further transforming temperature time series data into image sequences. Long Short-Term Memory network is constructed on the images to extract the depth features of the data to train the model. The experiments demonstrate that the deep learning prediction model containing the spatiotemporal information processing module and the deep learning prediction module is fully feasible in short-term regional temperature prediction. The comparison experiments show that the model proposed in this paper has better prediction results for classical models, such as convolutional neural networks and LSTM networks. The experimental conclusion shows that the method proposed in this paper can predict the distribution and change trend of temperature in the next 3 h and the next 6 h on a regional scale. The experimental result RMSE reached 0.63, showing high stability and accuracy. The model provides a new method for local regional temperature prediction, which can support the planning of production and life in advance and tend to save energy and reduce consumption.