Data Enhancement Method Research Articles

A stable diffusion enhanced YOLOV5 model for metal stamped part defect detection based on improved network structure

Machine-vision-based defect detection for large metal stamping is a fundamental requirement for improving product quality and inspection speed. However, the performance of machine vision in detecting defects is limited by the large variety of stamped products, significant dimensional differences, and long data-acquisition cycles. To overcome these problems, metal stamping detection has been achieved using the improved YOLOV5 model, which features a slim neck and a multiheaded self-attentive mechanism for detecting wrinkles, holes, and cracks in metal stamping. An image-processing module was used to reduce the impact of metal reflections and improve image quality. Stable diffusion improvement was used to augment the dataset to overcome the small dataset size problem and enhance its generalisation capability. In addition, a BotNet structure with a self-attentive mechanism was introduced into the YOLOV5 model backbone to improve the image feature extraction capability. We then optimised the prediction head structure of YOLOV5 to improve the detection speed and accuracy. Ablation experiments were performed to analyse and verify the effectiveness of each module. The results of the ablation experiments show that the mAP of our proposed stable diffusion improvement data enhancement method, and the YOLO-Bot-VOV algorithm, for metal stamped part defect detection reached 98.2 %, and the parameters were reduced by 0.432 million.

Disentangled variational auto-encoder enhanced by counterfactual data for debiasing recommendation

Recommender system always suffers from various recommendation biases, seriously hindering its development. In this light, a series of debias methods have been proposed in the recommender system, especially for two most common biases, i.e., popularity bias and amplified subjective bias. However, existing debias methods usually concentrate on correcting a single bias. Such single-functionality debiases neglect the bias-coupling issue in which the recommended items are collectively attributed to multiple biases. Besides, previous work cannot tackle the lacking supervised signals brought by sparse data, yet which has become a commonplace in the recommender system. In this work, we introduce a disentangled debias variational auto-encoder framework (DB-VAE) to address the single-functionality issue as well as a counterfactual data enhancement method to mitigate the adverse effect due to the data sparsity. In specific, DB-VAE first extracts two types of extreme items only affected by a single bias based on the collier theory, which are, respectively, employed to learn the latent representation of corresponding biases, thereby realizing the bias decoupling. In this way, the exact unbiased user representation can be learned by these decoupled bias representations. Furthermore, the data generation module employs Pearl’s framework to produce massive counterfactual data to help fully train the model, making up the lacking supervised signals due to the sparse data. Extensive experiments on three real-world data sets demonstrate the effectiveness of our proposed model. Specifically, our model outperforms the best baseline by 19.5% in terms of Recall@20 and 9.5% in terms of NDCG@100 in the best scenario. Besides, the counterfactual data can further improve DB-VAE, especially on the data set with low sparsity.

Defect detection of the surface of wind turbine blades combining attention mechanism

The proposed work introduces a novel, lightweight feature fusion network model based on the attention mechanism to address the issues of high time consumption and poor effectiveness in wind turbine blade surface defect detection. A bidirectional feature fusion network enhances the YOLOX network. The model employs an attention mechanism module to learn channel and spatial feature information. A classification loss function with an attenuation factor is designed to tackle the unbalanced distribution of learning weights for multiple samples in the blade surface defect detection task. Additionally, to mitigate recognition loss caused by target occlusion or overlap, the Soft-NMS method is utilized to eliminate redundant detection boxes. Sample imbalance is addressed by creating new samples using a geometric transformation-based multi-sample fusion data enhancement method and unsupervised learning MFF-GAN image editing techniques. A feature fusion network comparison test, classification loss function comparison test, and attention mechanism module position ablation test were completed using the blade surface defect dataset. The results indicate that the blade surface defect detection model incorporating the attention mechanism can effectively identify five types of defects with an average detection accuracy of mAP-0.5, achieving approximately 95.03 % and a detection frame rate of about 54.56 frame·s−1. Compared with the YOLOX-s network, YOLOv7-tiny network, and YOLOv8-s network, this model can improve the recognition rate and shorten the detection time, achieving high precision and rapid identification and localization of blade surface defects.

Bearing fault diagnosis method based on the Gramian angular field and an SE-ResNeXt50 transfer learning model

Fault diagnosis methods for rolling bearings based on deep learning have become a research hotspot. However, these methods mostly use convolutional neural networks (CNNs), which have the problem of gradient dispersion or disappearance as the network deepens. Moreover, directly converting vibration signals into images as network input cannot preserve the temporal correlation between signals. In the case of small datasets and complex and variable working conditions, the accuracy of fault diagnosis is low and the generalisation ability is poor. To solve the above problems, a rolling bearing fault diagnosis method based on the Gramian angular field (GAF) and an SE-ResNeXt50 transfer learning model is proposed. Firstly, the parameters of the GAF obtained from multiple experiments are selected and the one-dimensional time-series vibration signal is encoded by combining the data enhancement method, and converted into a Gramian angular difference field (GADF) diagram and a Gramian angular sum field (GASF) diagram with local time information and uniqueness. Then, a fine-tuning transfer learning strategy is used to transfer the pre-trained model parameters to an SE-ResNeXt50 model, which improves the training speed of the model and improves the overfitting problem of the model on small target datasets. Finally, the GAF diagram is used as the input to the model and a feature recalibration strategy is used to adaptively obtain the importance of each feature channel, which further improves the feature utilisation. To verify the effectiveness and superiority of the proposed method, the rolling bearing data from Case Western Reserve University are selected for experimental verification and the generalisation performance of the proposed method is tested under varying loads and different dataset scales. The results show that when there is only a small amount of data, the proposed method can still achieve high diagnosis accuracy for different loads and has better recognition accuracy and generalisation compared to other fault diagnosis methods.

One-Dimensional Convolutional Multi-branch Fusion Network for EEG-Based Motor Imagery Classification

The Brain-Computer Interface (BCI) system based on motor imagery (MI) is a hot research topic nowadays, which can control external devices through the brain and has a wide range of applications in rehabilitation, gaming, and entertainment. Due to the non-smooth, non-linear, and low signal-to-noise ratio of the MI EEG signal, it is challenging to accurately decode the MI task intention. A new end-to-end deep learning method is proposed to decode raw MI EEG signals without preprocessing, such as filtering and feature reinforcement. The 1D convolution is used to learn the time-frequency features in MI signals, and a four-branch fusion network is used as the main body to add a 1D CNN-AE block and 1D SE-block to enhance the algorithm's performance. Experiments on two publicly available datasets demonstrate that our proposed algorithm outperforms the current state-of-the-art methods. It achieves 86.11% and 89.51% on the BCI Competition IV-2a and the BCI Competition IV-2b datasets, respectively, and a 6.9% improvement in the generalizability test. The proposed data enhancement method can effectively alleviate the overfitting of the algorithm and improve the decoding performance. Further analysis shows that 1D convolution can effectively extract the features associated with the MI task.

Soybean Leaf Diseases Recognition Based on Generative Adversarial Network and Transfer Learning

Soybean leaf disease labeling data are not easy to obtain, and soybean leaf disease model training often needs a lot of data. Due to the limitations of fixed rules such as rotation and clipping, traditional data enhancement cannot generate images with diversity and variability. In view of the above problems, this study proposed a data enhancement method based on generative adversarial network to expand the original soybean leaf disease dataset. This method was based on cyclic confrontation network, and its discriminator uses dense connection strategy to realize feature reuse, so as to reduce the amount of calculation. In the training process, improved transfer learning is used to automatically fine tune the pre-training model. The accuracy of the optimized method in 9 kinds of soybean leaf disease image recognition is 95.84%, which is 0.98% higher than the traditional fine-tuning method. The experimental results show that this method based on generating confrontation network has significant ability in generating soybean leaf disease image, and can expand the existing dataset. In addition, this method also provides an effective data enhancement solution for the expansion of other crop disease image datasets.

Graph-Based Progressive Fusion Network for Multi-Modality Vehicle Re-Identification

Vehicle re-identification (Re-ID) is a critical task in intelligent transportation, aiming to match vehicle images of the same identity captured by non-overlapping cameras. However, it is difficult to achieve satisfactory results based on RGB images alone in darkness. Therefore, it is of great importance to consider multi-modality vehicle re-identification. Currently, the proposed works deal with different modality features through direct summation and fusion based on heat map, which however ignores the relationship between them. Meanwhile, there is a huge gap between the different modalities, which needs to be reduced. In this paper, to solve the above two problems, we propose a Graph-based Progressive Fusion Network (GPFNet) using a graph convolutional network to adaptively fuse multi-modality features in an end-to-end learning framework. GPFNet consists of a CNN feature extraction module (FEM), a GCN feature fusion module (FFM), and a loss function module (LFM). Firstly, in FEM, we employ a multi-stream network architecture to extract single-modality features and common-modality features and employ a random modality substitution module to extract mixed-modality features. Secondly, in FFM, we design an efficient graph structure to associate the features of different modalities and adopt a progressive two-stage strategy to fuse them. Finally, in LFM, we use GCN-aware multi-modality loss to constrain the features. For reducing modality differences and contributing better initial mixed-modality features to FFM, we propose random modality substitution as a data enhancement method for multi-modality datasets. Extensive experiments on multi-modality vehicle Re-ID datasets RGBN300 and RGBNT100 show that our model achieves state-of-the-art performance.

Overcoming the effect of muscle fatigue on gesture recognition based on sEMG via generative adversarial networks

Surface electromyography (sEMG)-controlled bionic prostheses have been extensively investigated recently. However, the majority of investigations pertaining to gesture recognition grounded in sEMG predominantly center their focus on the human body in a static and non-fatigued condition, thus neglecting the ramifications of muscle fatigue on the precision of gesture recognition. This study explores the effect of muscle fatigue on recognition accuracy in a gesture recognition task based on sEMG. The muscle fatigue induction experiment was designed, and eight subjects were recruited to participate in the experiment to collect the sEMG signal data sets of seven gesture actions under non-fatigue and fatigue conditions. Seven gesture actions under non-fatigue and fatigue conditions were identified by four classifiers, namely K-nearest neighbor (K-NN), support vector machine (SVM), decision tree (DT), and deep convolution neural network (CNN). The experimental results show that muscle fatigue has a great impact on the accuracy of gesture recognition. Specifically, using the four classifiers K-NN, SVM, DT and CNN trained in the non-fatigue state, the test accuracy of sEMG signals in the non-fatigue state is 96.7 %, 89.0 %, 87.3 % and 97.5 % respectively, while the test accuracy in the fatigue state is reduced to 53.3 %, 55.4 %, 45.8 % and 64.8 % respectively. In this regard, we propose a new method to overcome the effects of muscle fatigue, namely, the data enhancement method based on Wasserstein General Adversary Networks-Gradient Penalty (WGAN-GP), which is used to enhance the sEMG signal under fatigue. Through the data enhancement of the fatigue sEMG signal, the experimental results show that the final test accuracy in the fatigue state is improved by more than 20 %, which can reach 72.3 %, 80.9 %, 69.9 % and 92.1 % respectively. This shows that the method proposed by us can effectively overcome the influence of muscle fatigue on the accuracy of gesture recognition, and has made a great contribution to the improvement of the robustness of gesture recognition model based on sEMG signal.

Weld beads and defects automatic identification, localization, and size calculation based on a lightweight fully convolutional neural network

Weld quality inspection is a key aspect of modern manufacturing. Automatic identification, location, and dimensioning of welds and defects are necessary for complex industrial environments. Weld and defects dimensional data are important indicators for high precision assessment of weld quality, while precise positioning of welds and defects can provide important location data for subsequent processing. In this study, a novel weld and defects detection, location, and size calculation method is proposed based on deep fully convolutional neural networks. Firstly, 43 high-precision high-power laser weld plate scanned images are automatically cropped to 800 pixels × 800 pixels by the proposed sliding window sampling method for training, testing, and validation, while an image data enhancement method is proposed to quadruple the amount of data to improve the model performance. Next, the OU-net image semantic segmentation model is established to accurately identify different detection targets, which can achieve MIoU and MPA values of 78.15% and 85.18% on the validation set after sufficient training, respectively. Then, the proposed cropped segmentation image fusion recovery method is used to obtain the complete weld segmentation image for a complete display of the different inspection targets on the weld plate. Finally, a detection targets localization and size calculation method is applied to accurately obtain the position data and size data of different targets. The validation results show that the proposed detection strategy can locate and measure every weld seam, collapse, overlap, spatter, and hump on the weld plate.

GIS partial discharge data enhancement method based on self attention mechanism VAE-GAN

The reliability of geographic information system (GIS) partial discharge fault diagnosis is crucial for the safe and stable operation of power grids. This study proposed a data enhancement method based on a self-attention mechanism to optimize the VAE-GAN method and solve the problem of the lack of partial discharge samples and the unbalanced distribution between different defects. First, the non-subsampled contourlet transform (NSCT) algorithm was used to fuse the UHF and optical partial discharge signals to obtain a photoelectric fusion phase resolved partial discharge (PRPD) spectrum with richer information. Subsequently, the VAE structure was introduced into the traditional GAN, and the excellent hidden layer feature extraction ability of the VAE was used to guide the generation of the GAN. Then, the self-attention mechanism was integrated into the VAE-GAN, and the Wasserstein distance and gradient penalty mechanisms were used to optimize the network loss function and expand the sample sets to an equilibrium state. Finally, the KAZE and polar coordinate distribution entropy methods were used to extract the expanded samples. The eigenvectors of the sets were substituted into the long short-term memory (LSTM) network for partial discharge fault diagnosis. The experimental results show that the sample generation quality and fault diagnosis results of this method were significantly better than the traditional data enhancement method. The structure similarity index measure (SSIM) index is increased by 4.5% and 21.7%, respectively, and the average accuracy of fault diagnosis is increased by 22.9%, 9%, 5.7%, and 6.5%, respectively. The data enhancement method proposed in this study can provide a reference for GIS partial discharge fault diagnosis.

An Intelligent Diagnosis Approach Combining Resampling and CWGAN-GP of Single-to-Mixed Faults of Rolling Bearings Under Unbalanced Small Samples

Rolling bearing is a key component with the high fault rate in the rotary machines, and its fault diagnosis is important for the safe and healthy operation of the entire machine. In recent years, the deep learning has been widely used for the mechanical fault diagnosis. However, in the process of equipment operation, its state data always presents unbalanced. Number of effective data in different states is different and usually the gap is large, which makes it difficult to directly conduct deep learning. This paper proposes a new data enhancement method combining the resampling and Conditional Wasserstein Generative Adversarial Networks-Gradient Penalty (CWGAN-GP), and uses the gray images-based Convolutional Neural Network (CNN) to realize the intelligent fault diagnosis of rolling bearings. First, the resampling is used to expand the small number of samples to a large level. Second, the conditional label in Conditional Generative Adversarial Networks (CGAN) is combined with WGAN-GP to control the generated samples. Meanwhile, the Maximum Mean Discrepancy (MMD) is used to filter the samples to obtain the high-quality expanded data set. Finally, CNN is used to train the obtained dataset and carry out the fault classification. In the experiment, a single, compound and mixed fault cases of rolling bearings are successively simulated. For each case, the different sets considering the imbalance ratio of data are constructed, respectively. The results show that the method proposed significantly improves the fault diagnosis accuracy of rolling bearings, which provides a feasible way for the intelligent diagnosis of mechanical component with the complex fault modes and unbalanced small data.

Deep Learning Logging Sedimentary Microfacies via Improved U-Net

Well logging data contain abundant information on stratigraphic sedimentology. Artificial identification is usually strongly subjective and time-consuming. Pattern recognition algorithms like SVM may not adequately capture the depth-related variations in logging curve shape. This paper defines logging sedimentary microfacies as unidirectional 2D image segmentation and builds an improved U-net model to meet the requirements of logging sedimentary microfacies acquaintance. The proposed model contains three characteristics: (1) It removes pooling layers to avoid the loss of spatial features; (2) it utilizes multi-scale convolution blocks for mining multi-scale spatial features in logging data; (3) a one dimensional convolution layer is added to achieve deep single-direction segmentation. In this model, a 2D image composed of several standardized logging curves is used as the network’s input. In addition, we propose an effective data enhancement method and calculate the geometric feature attributes of well logging curves to reduce the complexity of the data characteristics. We tested the model on manually annotated validation datasets. Our method automatically measures fine sedimentary microfacies characteristics, improving the accuracy of sedimentary microfacies identification and achieving the desired result. Additionally, the model was tested on unlabeled actual logging data, which shows the generalizability of this deep learning method on different datasets.

A novel imbalanced dataset mitigation method and ECG classification model based on combined 1D_CBAM-autoencoder and lightweight CNN model

To address the problems of the imbalanced datasets, limited computing power, and memory capability of wearable devices, this paper proposes a new hybrid method for imbalanced dataset mitigation and a novel lightweight Convolutional Neural Network (CNN) framework for electrocardiogram(ECG) signal classification. First, data enhancement method is used by changing the amplitude of a small number of samples and combining with the weighted loss function in the deep learning model. Then, One-Dimensional Convolutional Block Attention Module (1D_CBAM) mechanism autoencoder model is proposed to compress the ECG heartbeat signals and extract the features of the ECG heartbeats. A modified one-dimensional lightweight CNN model based on MobileNet architecture is eventually used to categorize various arrhythmias utilizing the above features. Using the MIT-BIH arrhythmia dataset, the proposed model is tested by dividing the MIT Database(MITDB) into Dataset1 and Dataset2, in conjunction with suggestions from AAMI and other heartbeat signal classification methods. The experimental results show that the proposed imbalanced dataset mitigation method has higher performance than other methods in this paper, the proposed model has the most stable convergence and the lowest reconstruction loss (approximately 0.1) in the feature extraction process under the same conditions. During the classification process, the lightweight CNN model has approximately 4x fewer trainable parameters(Params: 12,788) and its performance [Dataset1: F1-Score: 96.24% ACC: 98.21%; Dataset2: F1-Score: 98.73% ACC: 99.61%] is comparable to the traditional CNN model [Dataset1: F1-Score: 96.98% ACC: 99.17%; Dataset2: F1-Score: 99.18% ACC: 99.72%].

Insulator Abnormal Condition Detection from Small Data Samples.

Insulators are an important part of transmission lines in active distribution networks, and their performance has an impact on the power system's normal operation, security, and dependability. Traditional insulator detection methods, on the other hand, necessitate a significant amount of labor and material resources, necessitating the development of a new detection method to substitute manpower. This paper investigates the abnormal condition detection of insulators based on UAV vision sensors using artificial intelligence algorithms from small samples. Firstly, artificial intelligence for the image data volume requirements was large, i.e., the insulator image samples taken by the UAV vision sensor inspection were not enough, or there was a missing image problem, so the data enhancement method was used to expand the small sample data. Then, the YOLOV5 algorithm was used to compare detection results before and after the extended dataset's optimization to demonstrate the expanded dataset's dependability and universality, and the results revealed that the expanded dataset improved detection accuracy and precision. The insulator abnormal condition detection method based on small sample image data acquired by the visual sensors studied in this paper has certain theoretical guiding significance and engineering application prospects for the safe operation of active distribution networks.

A Study of Contrastive Learning Algorithms for Sentence Representation Based on Simple Data Augmentation

In the era of deep learning, representational text-matching algorithms based on BERT and its variant models have become mainstream and are limited by the sentence vectors generated by the BERT model, and the SimCSE algorithm proposed in 2021 has improved the sentence vector quality to a certain extent. In this paper, to address the problem that the SimCSE algorithm has—that the greater the difference in sentence length, the smaller the probability that the sentence pairs are similar—an EdaCSE algorithm is proposed to perturb the sentence length using a simple data enhancement method without affecting the semantics of the sentences. The perturbation is applied to the sentence length by adding meaningless English punctuation marks to the original sentence so that the model no longer tends to recognise sentences of similar length as similar sentences. Based on the BERT series of models, experiments were conducted on five different datasets, and the experiments proved that the EdaCSE method improves an average of 1.67, 0.84, and 1.08 on the five datasets.

Software Subclassification Based on BERTopic-BERT-BiLSTM Model

With the continuous influx of application software onto the application software market, achieving accurate software recommendations for users in the huge software application market is urgent. To address this issue, each application software market currently provides its own classification tags. However, several problems still exist, such as the lack of objectivity, hierarchy, and standardization in these classifications, which in turn affects the accuracy of precise software recommendations. Accordingly, a customized BERTopic model is proposed to cluster the software description texts of the application software and the automatic tagging and updating of the application software tags are realized according to the clusters obtained by topic clustering and the extracted subject words. At the same time, a data enhancement method based on the c-TF-IDF algorithm is proposed to solve the problem of imbalance of datasets, and then the classification model based on the BERT-BiLSTM model is trained on the labeled datasets to classify the software in the dimension of the application function, so as to realize the accurate software recommendation for users. Based on the experimental verification of two datasets, 21 categories in the SourceForge dataset and 19 categories in the Chinese App Store dataset are subclassed by the clustering results of the customized BERTopic model, and the tags of 138 subclasses and 262 subclasses are formed, respectively. In addition, a complete tagged software description text dataset is constructed and the software tags are updated automatically. In the first stage of the classification experiment, the weighted average accuracy, recall rate, and F1 value can reach 0.92, 0.91, and 0.92, respectively. In the second stage, the weighted average accuracy, recall rate, and F1 value can all reach 0.96. After data enhancement, the weighted average F1 value of the classification model can be increased by up to two percentage points.

A fault diagnosis method for rolling bearings of wind turbine generators based on MCGAN data enhancement

In view of the problems such as poor diagnostic capability and generalization ability of wind turbine generator bearing fault diagnosis methods caused by complex wind turbine generator bearing conditions and few fault samples under actual operating conditions, a wind turbine generator bearing vibration signal data enhancement method based on improved multiple fully convolutional generative adversarial neural networks (MCGAN) was proposed. Firstly, two-dimensional time-frequency features are extracted from the raw data using a Short-Time Fourier Transform (STFT). Secondly, by incorporating multiple CGANs of different scales and a hybrid loss function, the original GAN network was enhanced to learn the intrinsic distribution of bearing vibration signals and generate diverse vibration signals with distinct bearing fault characteristics, resulting in an expanded dataset. Finally, a comparative experiment was conducted using real wind turbine generator-bearing data. The results demonstrate that the augmented samples generated by MCGAN contain rolling bearing fault information while maintaining sample distribution and diversity. By utilizing the augmented dataset to train commonly used fault diagnostic classifiers, the diagnostic accuracy for the original vibration signals exceeds 80%, providing a theoretical basis for addressing the scarcity of fault samples in practical engineering scenarios.

ERNIE-based Named Entity Recognition Method for Traffic Accident Cases

Traffic accident case named entity recognition, which helps mine key information in traffic accident texts, plays a vital role in downstream tasks such as the construction of knowledge graphs in road traffic and intelligent policing. In this paper, we construct a named entity recognition model based on the EDE (Entity Data Enhancement)-ERNIE-Bidirectional Gated Recurrent Unit Network (BiGRU)-Conditional Random Field (CRF) to address the current situation of low traffic accident case data and poor recognition of long-text entities. First, the amount of accident case data is enhanced using the entity random substitution method. Next, the text data of traffic accident cases are characterized as a dynamic word vector using the ERNIE pretraining model. Then, the BiGRU network learns the long-distance dependency relationship in the text to enhance the effect of the model on long-text entity recognition. Finally, the result sequence is constrained by the CRF layer to realize the named entity recognition model. The experimental part uses data related to real traffic accident cases in a domestic area. The data enhancement method increases the data volume three times compared to the original data volume. Experimental results show that the EDE-ERNIE-BiGRU-CRF model achieves better F1 values, recall and precision achieved better performance than the entity recognition methods of BERT-BiGRU-CRF, ERNIE-BiGRU-CRF, ERNIE-BiLSTM-CRF, ERNIE-CRF, ERNIE, BiGRU-CRF, ROBERTA-wwm-ext-BiGRU-CRF and verify its effectiveness for entity recognition in traffic accident cases.

Development of effective model for non-destructive detection of defective kiwifruit based on graded lines.

The accurate detection of external defects in kiwifruit is an important part of postharvest quality assessment. Previous studies have not considered the problems posed by the actual grading environment. In this study, we designed a novel approach based on improved Yolov5 to achieve real-time and efficient non-destructive detection of multiple defect categories in kiwifruit. First, a kiwifruit image acquisition device based on grading lines was developed to enhance the image acquisition. Subsequently, a kiwifruit dataset was constructed based on the external defect characteristics and a new data enhancement method was proposed to augment the kiwifruit samples. Thereafter, the SPD-Conv and DW-Conv modules were combined to improve Yolov5s, with EIOU as the loss calculation function. The results demonstrated that the improved model training loss value was 0.013 lower, the convergence was accelerated, the number of parameters was reduced, and the computational effort was increased. The detection accuracies of the samples in the test set, which included healthy, leaf-rubbing damaged, healed cuts or scarred, and sunburned samples, were 98.8%, 98.7%, 97.6%, and 95.9%, respectively, with an overall detection accuracy of 97.7%. The detection time was 8.0 ms, thereby meeting real-time sorting demands. The average detection accuracy and model size of SSD, Yolov5s, Yolov7, and Yolov5-Ours were compared. When the confidence threshold was 0.5, the detection accuracy of Yolov5-Ours was 10% and 6.4% higher than that of SSD and Yolov5s, respectively. In terms of the model size, Yolov5-Ours was approximately 6.5- and 4-fold smaller than SSD and Yolov7, respectively. Thus, Yolov5-Ours achieved the highest accuracy, adaptability, and robustness for the detection of all kiwifruit categories as well as a small volume and portability. These results can provide technical support for the non-destructive detection and grading of agricultural products in the future.

Machine-learning and water energy harvesting based wireless water consumption sensing system in buildings

In recent years, smart wireless remote water meters have become popular, but this kind of smart water meter needs an external power supply, and every 100 to 200 installed this type of water meter, needs to set up an independent collector. This kind of water meter is expensive, and the external power supply will increase the energy consumption of the building, which is not conducive to the circuit safety of the old community. In order to reduce the burden of manual meter reading and reduce the building's "carbon emissions". A wireless water consumption sensing system for buildings based on machine learning and water energy collection is designed. Images are collected and recognized by visual sensors to realize the visual reading of water meter data. The system is powered by a 12V micro water generator and transferred to lithium battery energy storage to realize the self-energy supply of the whole system. The lightweight YOLOv5 was used to implement the model training process. Then, on the premise of the small sample data set, the data set was expanded by a data enhancement method which randomly changed the image brightness and hue. Finally, the training model is deployed on MCU through the TensorFlow Lite framework to realize edge AI detection. Experiments show that the size of the lightweight YOLOv5 is reduced by 87.1% and the mAP reaches 92.3%. The system can complete the water meter wireless meter reading and realize the system self-power stably. In addition, the system can be expanded for other meter reading tasks based on visual features.