Electroencephalographic (EEG) signals exhibit stable and reproducible differences between estrus and non-estrus dairy cows, indicating that central nervous system activity reflects estrus-related regulation beyond external behavioral expression. Analysis across multiple EEG domains shows that estrus-associated neural modulation involves coordinated changes across frequency bands and recording channels and remains identifiable under varying behavioral conditions. Using multi-channel EEG recordings collected from 90 dairy cows, estrus detection was evaluated in a cross-subject evaluation, in which training and testing were conducted on data from different individuals. Under this evaluation protocol, the EEG-based approach achieved an accuracy of 92.7 f 1.5%, an area under the curve of 94.0 f 1.4%, a sensitivity of 94.1 f 1.4%, and a specificity of 95.4 f 1.3%, demonstrating robust discrimination across individuals rather than reliance on individual-specific neural patterns. Comparative model evaluation under unified experimental conditions indicates that appropriate feature representation and modeling strategies consistently capture estrus-related neural patterns. Attention-based interpretability analysis further reveals that EEG frequency bands contribute unequally to estrus discrimination, with low-frequency components, particularly the delta band, exhibiting the highest and most stable contribution across behavioral conditions. These results support EEG as a physiologically meaningful and interpretable information source for cross-individual estrus detection in dairy cows.

In unmanned aerial vehicle (UAV) flight behaviour understanding, the lack of a unified semantic representation for continuous multi-source temporal data makes it difficult to model flight events, behavioural relationships, and composite behaviours in an interpretable manner. To address this issue, this paper proposes a knowledge graph construction method for UAV flight behaviour understanding. The proposed method integrates atomic event detection, temporal knowledge modelling, and composite behaviour reasoning, thereby enabling the automatic transformation of continuous flight logs into structured behavioural knowledge. First, local statistical features, including smoothed velocity and robust climb rate, are extracted from multi-source flight logs. Adaptive dynamic decision boundaries based on local means and standard deviations are then used to detect atomic flight events such as Takeoff, Turn, Hover, and Descent. Next, a temporal knowledge graph is constructed with atomic events as the core semantic units to link UAV platforms, mission instances, and geographical regions. Finally, domain ontology and SWRL rules are introduced to support composite behaviour reasoning and semantic querying. Experiments conducted on a hybrid dataset derived from MUN-FRL and EuRoC MAV show that the proposed method achieves a macro-average F1 score of 0.83 across four typical event categories, with per-event F1 improvements ranging from 4 to 17 percentage points over the compared baselines under the same event-level evaluation protocol. Bootstrap-based validation provides additional evidence for the stability of the observed improvement under the current test setting. The constructed temporal flight knowledge graph contains more than 5600 entities. Typical semantic query response times remain within 30 ms under the current graph size and query setting. The proposed method provides an interpretable and structured framework for UAV flight behaviour analysis and understanding.

Estrus monitoring is essential for improving reproductive efficiency and optimizing artificial insemination timing in ewes. However, existing studies mainly focus on binary estrus detection and often rely on wearable sensors, invasive physiological measurements, or vision-based behavioral analysis methods, which are insufficient for fine-grained estrus stage discrimination and are limited by animal stress, high deployment cost, and sensitivity to farm environmental disturbances. To address these limitations, this study proposes a non-contact multimodal framework for classifying ewe estrus stages into diestrus, estrus, and metestrus. Thermal infrared features were extracted using an improved ResNet18 incorporating initial convolution (INC), squeeze-and-excitation (SE), and multi-scale fusion (MSF) modules, denoted as INC-SE-MSF-ResNet18. In parallel, bilateral eye temperature features and environmental temperature and humidity features were extracted by multilayer perceptron (MLP)-based eye temperature feature extractor and MLP-based environmental variable feature extractor. These three types of features were then concatenated into a multimodal feature and classified using a Light Gradient Boosting Machine (LightGBM). Under the same LightGBM classifier, INC-SE-MSF-ResNet18 generated more discriminative thermal infrared features than ResNet18, MobileNetv3, EfficientNet, ShuffleNet, and DenseNet, yielding improvements of up to 15.10% in recall, 7.67% in F1-score, and 3.10% in accuracy. Under the same feature extraction and fusion strategy, LightGBM consistently outperformed support vector machine (SVM), eXtreme Gradient Boosting (XGBoost), gradient boosting decision tree (GBDT), random forest (RF), decision tree (DT), logistic regression (LR), and convolutional neural network (CNN) across all estrus stages, with precision improvements of up to 21.23% over CNN, 13.17% over DT, and 11.11% over RF. Temperature analysis further revealed clear stage-dependent divergence between bilateral eye temperatures and rectal temperature (RT). The bilateral eye temperatures peaked during estrus, whereas rectal temperature decreased during estrus and increased markedly during metestrus. These results indicate that surface and core temperatures reflect different physiological processes and exhibit unsynchronized temporal dynamics across the estrous cycle. Overall, this study extends infrared thermography-based ewe estrus monitoring from conventional binary detection to protocol-defined estrus stage classification, providing a feasible approach for ovulation timing inference and precision reproductive management in large-scale sheep farming systems.

This paper proposes a memristor-based artificial synapsis-like CMOS neuromorphic circuit with achieving Pavlov’s associative memory learning and forgetting functions. It consists of four key functional modules and can adaptively identify danger/temptation signals and then trigger the corresponding behavioral responses. The overall neuromorphic system is compacted into a single integrated circuit (IC) utilizing compact two-stage operational amplifiers (OpAmps). Large-size output transistors are placed by multi-finger layouts to ensure sufficient driving capability while maintaining a minimized layout area, with the smaller circuit dimension of only 0.011 mm2 layout area, and the lower power consumption of only 37.38 mW. 180 nm/1.8 V standard BCD process is adopted to implement the circuit design and performance simulation on Cadence. The execution results demonstrate that, the weight value of the proposed memristor-based artificial synaptic can be accurately adjusted in training process, and further the whole human-like neuromorphic system can effectively imitate the Pavlov’s conditioned reflex such as associative memory learning and forgetting process.

This study aimed to investigate the effects of hydroxyapatite (HA) nanoparticles and cryoprotectant concentrations (CPAs) on ultrastructure, developmental competence and gene expression in bovine metaphase II (MII) oocytes after vitrification. Bovine MII oocytes were vitrified in vitrification solution supplemented with HA nanoparticles sizes (20, 40, or 60nm) and concentrations (0.01, 0.05, or 0.1%), and the optimal conditions of HA nanoparticles were determined based on survival rate, mitochondrial membrane potential (MMP), and developmental competence. Subsequently, we investigated the synergistic effects of HA and CPAs on ultrastructure, MMP, developmental competence, and related genes expression in bovine MII oocytes. The results showed that 40nm 0.05% HA group significantly enhanced MMP levels, cleavage and blastocyst rates compared to the control group (P < 0.05). The VS1-HA group (VS1-HA: vitrification solution containing 17.5% EG (v/v) and 17.5% DMSO (v/v) with HA) exhibited the highest cleavage rates (43.67%) and blastocyst rates (15.19%), further reduced ultrastructural damage including mitochondrial swelling, cristae disruption, and microvilli loss, compared with the VS-HA group (VS-HA: vitrification solution containing 20% EG (v/v) and 20% DMSO (v/v) with HA). Furthermore, the relative mRNA levels of mitochondrial genes ND5 and ATPase8, and zona pellucida gene ZP3 (P < 0.05) were significantly higher than those in the VS group, but no difference in KRT8 or COX1 (P > 0.05). In conclusion, HA nanoparticles could reduce the negative effects of cryoinjury on the ultrastructure of certain organelles and the mRNA levels of specific related genes, thereby improving the developmental competence of vitrified bovine MII oocytes.

Although rectal temperature is reliable, its measurement requires manual handling and causes stress to animals. IRT provides a non-contact alternative but often ignores bilateral eye temperature differences. This study presents an E-S-YOLO11n model for the automated detection of the binocular regions of sheep, which achieves remarkable performance with a precision of 98.2%, recall of 98.5%, mAP@0.5 of 99.40%, F1 score of 98.35%, FPS of 322.58 frame/s, parameters of 7.27 M, model size of 3.97 MB, and GFLOPs of 1.38. Right and left eye temperatures exhibit a strong correlation (r = 0.8076, p < 0.0001), However, the eye temperatures show only very weak correlation with rectal temperature (right eye: r = 0.0852; left eye: r = −0.0359), and neither figure reaches statistical significance. Rectal temperature is 7.37% and 7.69% higher than the right and left eye temperatures, respectively. Additionally, the right eye temperature is slightly higher than the left eye (p < 0.01). The study demonstrates the feasibility of combining IRT and deep learning for non-invasive eye temperature monitoring, although environmental factors may limit it as a proxy for rectal temperature. These results support the development of efficient thermal monitoring tools for precision animal husbandry.

Electricity theft is an essential issue within non-technical losses of energy. Its complexity intensifies with the emergence of novel network attacks. Previous studies have highlighted the efficacy of machine learning (ML) techniques in detecting individual instances of electricity theft using raw data. However, insufficient depth of data utilization, inadequate detection of compound attacks, and poor interpretability of models limit their reliability. To overcome these predicament, electrical and environmental features are obtained through manual curation, and their relationship and temporal lag effects are evaluated using Pearson correlation coefficients. Moreover, a weighted average electricity theft detection algorithm based on Deep and Cross Network Version 2 (DCN V2)-Transformer is proposed. The independent training and configuration optimization of the DCN_V2 and Transformer are implemented by grid search methodology, respectively. During the model integration, the optimized predictions yielded by the two models are adeptly combined according to appropriate weights. Furthermore, the decision-making process of Transformer is analyzed using visual attention weight methodology. The experimental results demonstrate a significant leap in the performance of the proposed model by considering environmental factors. The evaluation metrics perform remarkably, with an accuracy of 95.9%, precision of 94.8%, recall of 96.8%, Fl score of 95.8%, and false positive rate of 5%. Additionally, the comparative analysis verifies the advantages of this approach over convolutional neural networks (CNN), long short-term memory network (LSTM), CNN+LSTM, Wide+DeepCNN, support vector machine (SVM), and decision tree+SVM algorithms. This study provides a promising technique for electricity theft detection and assists in understanding the operating mechanism of detection model.

The combination of multimodal features based on vocalization and behavioral traits has been proven to enhance the robustness and accuracy of estrus detection. However, challenges still need to be addressed to improve the reliability and practicality of the identification of estrus cows, including unclear association mechanisms between multi-feature estrus traits and complex estrus states, inadequate feature selection and fusion strategies, and limitations in algorithm performance. To cope with these difficulties, the Friedman test, Mantel test, Spearman rank correlation coefficient, Kruskal-Wallis test, and Canonical Correspondence Analysis (CCA) were employed to explore the complex interactions among high-dimensional estrus data. Moreover, the model interpretability framework based on the self-attention mechanism was constructed to reveal the importance of critical features and optimize feature combinations. In addition, the multimodal fusion approach integrating standardization processing, principal component analysis, Fourier transform, statistical indicators extraction, and semi-tensor product was designed to elevate the depth of representation and the fusion effect of multidimensional data by comprehensively analyzing the relationships among various features. Furthermore, a neural network-optimized Hidden Markov Model (NN-HMM) for estrus detection was proposed to promote the capability of estrus detection by overcoming the imperfections of traditional HMM in capturing long distance dependence effect in state transition matrices, describing complex feature relationship in emission probability matrices, and dynamic adaptability of optimal path generation. The experimental results demonstrated that the selection of optimal feature combination integrating acoustic and behavioral features (number of bellowings, number of lowings, number of consecutive bellowings, vocalization frequency, standing duration, lying duration, walking duration, feeding duration, activity index, number of standing mounts, number of social behaviors, and ruminating variation index) improved estrus detection accuracy by over 18 % compared to using suboptimal feature combinations. Meanwhile, compared with Support Vector Machine (SVM), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) models, the accuracy of the developed multimodal fusion-based detection method increased by 8.8 %, 7.4 %, and 6.1 %, while the precision enhanced by 7.4 %, 6.0 %, and 4.3 %, respectively. Blind testing conducted on multisource estrus data from 24 multiparous and 16 primiparous cows found that the proposed prediction method advanced the prediction of estrus onset by up to 85 and 40 mins for primiparous cows and up to 75 and 60 mins for multiparous cows, respectively, significantly surpassing conventional activity index and acoustic-based methods. Therefore, the method proposed in this study undoubtedly provides a reliable solution for timely and efficient detection of estrus in dairy farms.

Eletheft, as a core issue of non-technical losses, has become more complex with the emergence of new network attack methods. Studies indicate that machine learning models relying solely on raw data have limitations when handling complex and covert composite attacks. In response, this study undertakes a comprehensive approach, proposing an ensemble learning-based DCN-Transformer weighted algorithm to address this challenge. We begin by manually extracting electricity consumption and environmental features and use Pearson correlation to quantify their relationships and time-lag effects. The DCN and Transformer models are then independently trained and optimized, with key parameters fine-tuned using grid search to enhance detection performance. Finally, during the integration phase, the prediction results of the two models are intelligently weighted and merged to construct a comprehensive prediction score. Our experimental results, which are the product of this thorough approach, show that considering environmental factors improves model performance, achieving an accuracy of 93.9%, precision of 92.8%, recall of 93.8%, F1 score of 91.8%, and a false detection rate of only 6%. Compared to other algorithms, our proposed method demonstrates significant advantages, further validating the thoroughness of our study.

Pre-trained models trained with internet-scale data have achieved significant improvements in perception, interaction, and reasoning. Using them as the basis of embodied grasping methods has greatly promoted the development of robotics applications. In this paper, we provide a comprehensive review of the latest developments in this field. First, we summarize the embodied foundations, including cutting-edge embodied robots, simulation platforms, publicly available datasets, and data acquisition methods, to fully understand the research focus. Then, the embodied algorithms are introduced, starting from pre-trained models, with three main research goals: (1) embodied perception, using data captured by visual sensors to perform point cloud extraction or 3D reconstruction, combined with pre-trained models, to understand the target object and external environment and directly predict the execution of actions; (2) embodied strategy: In imitation learning, the pre-trained model is used to enhance data or as a feature extractor to enhance the generalization ability of the model. In reinforcement learning, the pre-trained model is used to obtain the optimal reward function, which improves the learning efficiency and ability of reinforcement learning; (3) embodied agent: The pre-trained model adopts hierarchical or holistic execution to achieve end-to-end robot control. Finally, the challenges of the current research are summarized, and a perspective on feasible technical routes is provided.

Routing protocols, as a crucial component of the internet of things (IoT), play a significant role in data collection and environmental monitoring tasks. However, existing clustering routing protocols suffer from issues such as uneven network energy consumption, high communication delays, and inadequate adaptation to topology changes. To address these issues, this study proposes an adaptive routing algorithm to balance energy consumption and delay using game theory and deep Q-network (DQN) algorithms (EDRP-GTDQN). Specifically, EDRP-GTDQN evaluates the importance of node positions using node centrality and integrates a game-theoretic-based approach to select optimal cluster heads in terms of node centrality and residual energy. Moreover, graph convolutional networks (GCN) and DQN are incorporated to construct transmission paths for cluster heads, adapt to network topology changes, and balance energy consumption and performance. Furthermore, a cluster rotation mechanism is employed to optimize overall network energy consumption and prevent the formation of hotspots. Experimental results demonstrate that EDRP-GTDQN achieves average performance improvements of 19.76%, 30.04%, 44.2%, and 61.42% in average energy consumption, network lifetime, and average end-to-end delay compared to conventional routing protocols such as EECRAIFA, MRP-GTCO, DEEC, and MH-LEACH. Therefore, EDRP-GTDQN is undoubtedly an effective solution to reduce energy consumption and enhance service quality in wireless sensor networks.

This article presents a novel CMOS nested chopper instrumentation amplifier (NCIA) suitable for physiological signal (such as EEG, ECG and EMG) acquisition systems. By incorporating a DC offset suppression module and impedance boosting loop, the implemented instrumentation amplifier achieves high input impedance and low DC offset voltage. The usage of novel nested chopping technology with noise shifting in the filtering circuit effectively eliminates low-frequency noise, which makes it well-suited for analog front-end (AFE) sensing systems for precision extraction of weak physiological signals. The proposed NCIA circuit is implemented based on 180 nm/1.8 V standard BCD technology. Under 1.8V power supply voltage, the power consumption of the overall amplifier circuit is 4.17 mu W, the total input reference noise is 771.365 nVrms, and the total circuit layout area is 5.47 x 10- 2 mm2.

Infrared thermography for non-contact measurement of eye temperatures effectively reduces sheep stress, lowers labor costs, and improves detection efficiency. This study aims to evaluate the practicability and feasibility of replacing manual annotation with an improved YOLOv7 automated detection method for sheep eye temperature. The proposed SE-SC-YOLOv7 algorithm achieves a precision of 99.5 %, recall of 99.3 %, FPS of 99.3, AP0.5 of 99.7 %, and mAP0.5:0.95 of 73.3 %. This mAP0.5:0.95 outperforms Centernet, Faster-RCNN, SSD, and YOLOx by 7.32 %, 30.20 %, 11.40 %, and 7.79 %, respectively. The mean absolute error of the maximum eye temperature between the SE-SC-YOLOv7 algorithm and manual annotation is 0.033 degrees C, which is 16.24 %, 60.29 %, 6.54 %, and 18.58 % lower compared to the Centernet, Faster-RCNN, SSD, and YOLOx algorithms, respectively. Additionally, Spearman correlation analysis shows that the maximum, minimum, and average eye temperatures detected by the SE-SC-YOLOv7 algorithm are highly correlated with the manually annotated temperatures (rho = 0.989, P < 0.001. rho = 0.922, P < 0.001. rho = 0.963, P < 0.001). The paired samples Wilcoxon signed-rank test demonstrates no significant difference between the eye temperatures detected by the SE-SC-YOLOv7 algorithm and manual annotation. The Bland-Altman consistency test proves that at the 95 % significance level, the SE-SC-YOLOv7 algorithm can reliably replace manual annotation. Therefore, the proposed method undoubtedly offers a promising solution for effectively detecting sheep eye temperature.

False Data Injection Attacks (FDIA) significantly threaten the state estimation of cyber-physical systems (CPS) in the power grid, jeopardizing the secure operation of the electricity system. Currently, most FDIA detection methods are designed for direct current systems, and traditional DC FDIA detection methods cannot effectively detect alternating current FDIA with satisfactory performance. The existing AC FDIA detection methods typically employ time-series measurement data for direct detection, leading to suboptimal detection accuracy. Moreover, these methods focus on detecting the presence of attacks, failing to locate the positions affected by the attacks accurately. To address these issues, this study introduces a deep learning detection model based on Short-Time Fourier Transform (STFT) for the localization of stealthy FDIA in AC systems. Initially, low-quality data affected by FDIA is removed using a BDD detector. Then, the filtered measurement data is transformed from the time domain to the time-frequency domain using STFT. Finally, the results of the STFT are input into a dual-channel convolutional neural network (DCNN) to extract time-frequency domain features for training and learning, aiming to locate the position of the FDIA attack. Experiments conducted in the IEEE 14 and IEEE 118 bus systems demonstrate that the detection method proposed in this article achieves an average AUC value that is respectively 0.0465, 0.0812 and 0.0255 higher than those of the STFT+BPNN, STFT+LSTM, and STFT+Bi-LSTM algorithms. Furthermore, under various noise intensities and attack strengths, it maintains accuracy, recall, precision, F1 score, and RACC at levels above 0.9036, 0.9425, 0.9588, 0.9506, and 0.899, respectively.

Acoustic technology has great application prospects in aquaculture. In particular, two indispensable, critical technologies for the future aquaculture industry are multi-sensor acquisition that can achieve multi-scale information fusion, collection and establishment of a global acoustic fish database and highly developed deep learning intelligent algorithms that can establish a correlation mechanism between fish acoustic and behaviour characteristics. Acoustic technology offers remarkable advantages in large and turbid water bodies for studying spatial and temporal distribution patterns of aquatic organism populations, developing on-demand feeding systems and estimating biomass. This article reviews the development of acoustic technology and its application in aquaculture over the last 30 years. It further analyses, in detail, the advantages and disadvantages of acoustic technology in evaluating aquatic organism biomass and morphological and physical indicators, aquatic organism behaviour and welfare improvement. Challenges of acoustic technology in acquiring dynamic target data accurately, building a global acoustic fish database and establishing connections between fish behaviour and acoustic characteristics are also discussed. In brief, this article aims to help researchers and practitioners better understand the current state-of-the-art acoustic technologies, which can provide strong support for smart aquaculture applications.

A power-efficient, wide-band, programmable pseudo-differential ring-oscillator charge-pump phase-locked loop (CP-PLL) is proposed. The ring-VCO, characterized by its low power consumption and broadband, is achieved based on a feedforward pseudo-differential configuration. The linearity of ring-VCO is improved by using the source negative feedback topology. Through the rail-to-rail operatingamplifier clamping current source, the current matching accuracy is improved to realize a high-performance CP circuit. The results show that the output frequency range of the phase-locked loop is 0.6-6 GHz at a 1.2 V supply voltage, and the power consumption is 1.372 mW at 5 GHz with a lock-up time of 1.72 mu s and RMS jitter of 9.3 ps. The power consumption is as low as 0.643 mW at 2.5 GHz and 1.067 mW at 4 GHz, and the final layout area is 0.00716 mm2. 2 . The implemented CPPLL can be used effectively in wireless RF communication system of NB-IoT and intelligent edge computing scenario.

False data injection attacks (FDIA) exploit the vulnerabilities of bad data detection in energy management systems to maliciously tamper with state estimation results, seriously jeopardizing the safe and reliable operation of power systems. In order to promptly detect FDIA, recent studies have used machine learning techniques to extract attack characteristics and detect FDIA based on changes in these characteristics. Current research predominantly focuses on detecting a single type of FDIA attack model and topology. However, the increasing diversification of FDIA construction methods and the variability of topological structures in power systems can reduce or even invalidate detection effectiveness. To cope with the abovementioned difficulties, this study introduces a multimodal deep learning detection model based on variational graph auto-encoders (VGAE), temporal convolutional networks (TCN), and gated recurrent units (GRU). The topological features of power systems are obtained through VGAE to adapt to the detection needs of various topological structures and performance enhancement. These features are then integrated with preprocessed measurement data via BDD to form multimodal data. Furthermore, this multimodal data is used to locate and classify FDIA with complete information, FDIA with incomplete information, and topology attack after applying a multi-label classification algorithm based on TCN-GRU for temporal feature extraction. Experiments carried out on the IEEE 14 and IEEE 118 bus systems show that the proposed method is robust and has high detection performance. The results indicate that the proposed detection method achieves AUC values that are, on average, 0.085, 0.145, and 0.04 higher than those of CNN, LSTM, and CNN-LSTM, respectively. Moreover, under various noise and attack intensities, recall, precision, F1 score, and RACC remain above 0.859, 0.877, 0.867, and 0.818, respectively, with a classification accuracy greater than 0.912. This study provides a unique perspective on detecting FDIA across various attack models and power system topologies.

Recently, the detection difficulty of emerging power grid cyberattacks for electricity theft far surpasses that of traditional physical power theft behavior. Reliable cyberattack detection is of great importance for ensuring stable smart grid operation and protecting the interests of power supply companies. Recently, a few studies have demonstrated acceptable performance in detecting individual or minority attacks using machine learning. However, difficulties such as poor data preprocessing, unwarranted feature selection, and deficiency of joint attack detection have limited their reliability. To unravel these issues, this study uses a long short-term memory (LSTM)-based multi-step prediction model to forecast and fill in missing data. Moreover, a series of statistical methods are utilized to explore optimal feature selection by analyzing the significance, correlation, and applicability between normal and attack feature data. Furthermore, a federated-learning-based stacking ensemble gate recurrent unit algorithm (FL-SE-GRU) is proposed for electricity theft detection and cyberattack classification, and its effectiveness is verified by comparison with existing methods. The results of experiments show that the LSTM-based multi-step prediction model exhibits a remarkable data interpolation effect. Meanwhile, FL-SE-GRU achieves 95.0% accuracy, 96.6% precision, 93.8% sensitivity, and 95.1% F1 score in detecting electricity theft, and reaches 96.8% accuracy, 96.1% precision, 97.4% sensitivity, and 96.7% F1 score in classifying 9 kinds of cyberattacks, respectively. This study provides unique perspectives to cope with the increasingly complicated cyberattacks for electricity theft.

The stability of power systems is paramount to industrial operations. The deleterious inherent characteristics of false data injection attacks (FDIA) have drawn substantial interest due to their severe threats to power grids. Contemporary detection systems face numerous challenges as attackers employ various tactics, such as injecting complex elements into measurement data and formulating quick attack strategies against critical nodes and transmission lines in the power grid network topology. Conventional models often fail to adapt to the intricacies of practical situations because they focus predominantly on detecting individual components. To overcome the above predicaments, this paper proposes a lightweight detection model integrating deep separable convolutional layers, squeeze neural networks, and a bidirectional long short-term memory architecture named DSE-BiLSTM. The acquisition process of network topological characteristics is accomplished through variable graph attention autoencoder (VGAAE). This approach leverages the effectiveness of the graph convolution (GCN) layer to acquire each node’s topological feature and the graph attention (GAT) module to identify and extract the topological features of critical nodes. Furthermore, the topology information obtained by the both techniques is embedded in one-dimensional vector space in the same form as measurement data. By combining the output of VGAAE with meter measurements, the feature fusion of temporal and spatial modalities is realized. DSE-BiLSTM with optimal hyperparameters achieves an F1-score of 99.56% and a row accuracy (RACC) of 93.10% on the conventional dataset. The experimental results of FDIA detection with composite datasets of IEEE 14-bus and IEEE 118-bus systems show that the F1-score and RACC of DSE-BiLSTM remain above 84.51% and 83.56% under various attack strengths and noise levels. In addition, as the power grid network scales up, noise level’s effect on detection performance decreases, while attack strength’s effect on recognition capability increases. DSE-BiLSTM can effectively process the composite data of spatiotemporal multimodes and provides a feasible solution for the localization and detection of FDIA in realistic scenes.

Timely identification of cows in heat is a fundamental issue in dairy farming. Although current studies have confirmed that several cow vocalisation characteristics can be used for oestrus identification, the reliability and practicability of oestrus detection methods are limited due to background noise, a lack of data support for sound feature selection, and the unsatisfactory effectiveness of vocalisation recognition algorithms.To overcome these predicaments, a dual-channel recording device and a sound event extraction method were developed in this study. The experimental results in respect of 62 Holstein dairy cows showed that the sound sample recording accuracy was 80.4%, and the noise filtering accuracy was 94.3%, indicating the effectiveness of the vocalisation extraction algorithm in a noisy farm environment. Moreover, two unique sound features with remarkable discrimination ability, i.e., the number of consecutive vocalisations and the maximum consecutive times, were used to improve the effectiveness of oestrus detection. The Friedman test, the Spearman rank order correlation coefficient, and the Kruskal-Wallis test were used to obtain the combination of features with the most optimal discrimination. Subsequently, a dual-LSTM (Long Short-Term Memory) joint discriminant strategy based on optimal combinations was proposed to promote oestrus detection performance. The results of the blind test on the vocalisation data of 20 oestrus cows and 11 non-oestrus cows demonstrated that the oestrus detection rate of the discrimination strategy reached 100%, and had a temporal advantage over the activity-index-based method in determining the onset of oestrus.The improvements of the selection and-based oestrus detection method mainly included accurately extracting individual cow vocalisations, supplementing the explicability of feature selection, and enhancing the refinement of oestrus recognition. Meanwhile, it is of positive significance for applying sound acquisition devices in oestrus detection. Furthermore, combining sound identification with other automated detection techniques appears to be a promising method for promoting the oestrus detection rate.