A MACHINE VISION SYSTEM FOR INSPECTING WOOD SURFACE DEFECTS BY USING NEURAL NETWORK

With the development of wood industry, the processing of wood products becomemore significant. This paper discusses the developmen of machine vision system used to inspect andclassny the various types of defects of wood suxface. The surface defeds means the variations ofcolour and textUre. The machine vision system is to dated undesirable 'defecs' that can appear onthe surface of rough wood lwnber. A neural network was used within the Blackboard framework fora labeling verification step of the high-level recognition module of vision system. The system hasbere successfully tested on a number of boards from several different species.

Neural Network in Computer Vision for RoboCup Middle Size League

Robot World Cup Initiative (RoboCup) is a worldwide competition proposed to advance research in robotics and artificial intelligence. It has a league called RoboCup soccer devoted for soccer robots, which is a challenge because robots are mobile, fully autonomous, multi-agents, and they play on a dynamic environment. Moreover, robots must recognize the game entities, which is a crucial task during a game. A camera is usually used as an input system to recognize ball, opponents, soccer field, and so on. These elements may be recognized applying some tools of computational intelligence, for example an artificial neural network. This paper describes the application of an artificial neural network on middle size robotic football league, where a multilayer perceptron neural network is trained with the backpropagation algorithm, to classify elements on the image. Each output neuron represents an entity and its output value depends on the current entity that is present on the image. The results show that an artificial neural network successfully classified the entities. They were recognized even when similar color entities were present on the image.

Automatic Identification of Tomato Maturation Using Multilayer Feed Forward Neural Network with Genetic Algorithms (GA)

We set up computer vision system for tomato images. By using this system, the RGB value of tomato image was converted into HIS value whose H was used to acquire the color character of the surface of tomato. To use multilayer feed forward neural network with GA can finish automatic identification of tomato maturation. The results of experiment showed that the accuracy was up to 94%.

基于改进Vision Transformer网络的农作物病害识别方法

基于DCNN模型的农作物病害识别方法在实验室环境下识别准确率高,但面对噪声时缺少鲁棒性.为了兼顾农作物病害识别的精度和鲁棒性,本文在标准ViT模型基础上加入增强分块序列化和掩码多头注意力,解决标准ViT模型缺乏局部归纳偏置和视觉特征序列的自注意力过于关注自身的问题.实验结果表明,本文的EPEMMSA-ViT模型对比标准ViT模型可以更高效的从零学习;当添加预训练权重训练网络时,EPEMMSA-ViT模型在数据增强的PlantVillage番茄子集上能够得到99.63%的分类准确率;在添加椒盐噪声的测试数据集上,对比ResNet50、DenseNet121、MobileNet和ConvNeXt的分类准确率分别提升了6.08%、9.78%、29.78%和12.41%;在添加均值模糊的测试数据集上,对比ResNet50、DenseNet121、MobileNet和ConvNeXt的分类准确率分别提升了18.92%、31.11%、20.37%和19.58%.

Overview of Object Detection Algorithms Using Convolutional Neural Networks

In today’s world, computer vision technology has become a very important direction in the field of Internet applications. As one of the basic problems of computer vision, object detection has become the basis of many vision tasks. Whether we need to realize the interaction between images and text or recognize fine categories, it provides reliable information. This article reviews the development of object detection networks. Starting from RCNN, we introduce object detection based on candidate regions, including Fast R-CNN, Faster R-CNN, etc.;and then start to introduce single-shot networks including YOLO, SSD, and Retina Net, etc. Detectors are the most excellent methods at present. By reviewing the current research status of object detection networks, it provides suggestions for the further development trend and research of object detection.

Convolutional Neural Network and Bayesian Gaussian Process in Driving Anger Recognition

With the development of motorization, road traffic crashes have become the leading cause of death in many countries. Among roadway traffic crashes, almost 90% of accidents are related to driver behaviors, wherein driving anger is one of the most leading causes to vehicle crash-related conditions. To some extent, angry driving is considered more dangerous than typical driving distraction due to emotion agitation. Aggressive driving behaviors create many kinds of roadway traffic safety hazards. Mitigating potential risk caused by road rage is essential to increase the overall level of traffic safety. This paper puts forward an integrated computer vision model composed of convolutional neural network in feature extraction and Bayesian Gaussian process in classification to recognize driver anger and distinguish angry driving from natural driving status. Histogram of gradients (HOG) was applied to extract facial features. Convolutional neural network extracted features on eye, eyebrow, and mouth, which are considered most related to anger emotion. Extracted features with its probability were sent to Bayesian Gaussian process classier as input. Integral analysis on three extracted features was conducted by Gaussian process classifier and output returned the likelihood of being anger from the overall study of all extracted features. An overall accuracy rate of 86.2% was achieved in this study. Tongji University 8-Degree-of-Freedom driving simulator was used to collect data from 30 recruited drivers and build test scenario.

Color Reproduction on CRT Displays via BP Neural Networks Under Office Environment

A CRT characterization method based on color appearance matching is presented. A matching between Munsell color chips and CRT charts was obtained in vision perceiver in typical office environment and viewing condition by recommending. And neural networks were utilized to accomplish the color space conversion from CIE standard color space to CRT device color space. The neural networks related the color space conversion and color reproduction of soft/hard-copy directly to the influence of the illuminance and viewing condition in vision perceiver. The average color difference of training samples is 3.06 and that of testing samples is 5.17. The experiment results indicated that the neural networks can satisfy the requirements for the color appearance of hard-copy reproduction in CRT.

AN IMAGE SEGMENTATION APPROACH BASED ON FUZZY-NEURAL-NETWORK HYBRID SYSTEM

This paper presents a new solution to the image segmentation problem, which is based on fuzzy-neural-network hybrid system (FNNHS). This approach can use the experiential knowledge and the ability of neural networks which learn knowledge from the examples, to obtain the well performed fuzzy rules. Furthermore this fuzzy inference system is completed by neural network structure which can work in parallel. The segmentation process consists of pre-segmentation based on region growing algorithm and region merging based on FNNHS. The experimental results on the complicated image manifest the utility of this method.

Fully Connected Convolutional Neural Network in PCB Soldering Point Inspection

In Electronics Manufacturing Services (EMS) industry, Printed Circuit Board (PCB) inspection is tricky and hard, especially for soldering point inspection due to the extremely tiny size and inconsistent appearance for uneven heating in reflow soldering process. Conventional computer vision technique based on OpenCV or Halcon usually cause false positive call for originally good soldering point on PCB because OpenCV or Halcon use the pre-defined threshold in color proportion for deciding whether the specific soldering point is OK or NG (not good). However, soldering point forms are various after heating in reflow soldering process. This paper puts forward a VGG structure deep convolutional neural network, which is named SolderNet for processing soldering point after reflow heating process to effectively inspect soldering point status, reduce omission rate and error rate, and increase first pass rate. SolderNet consists of 11 hidden convolution layers and 3 densely connected layers. Accuracy reports are divided into OK point recognition and NG point recognition. For OK soldering point recognition, 92% is achieved. For NG soldering point recognition, 99% is achieved. The dataset is collected from KAGA Co. Ltd Plant in Suzhou. First pass rate at KAGA plant is increased from 25% to 80% in general.

基于Vision Transformer和迁移学习的垃圾图像分类研究

为解决垃圾图像分类中分类准确率低及小样本类别性能差的问题,以生活垃圾图像为研究对象,以正确识别生活垃圾类别为研究目标,利用Vision Transformer模型为分类网络架构,使用迁移学习机制实现该模型在华为云垃圾分类数据集上的训练及分类推理。实验结果表明,基于注意力机制的分类模型相较于基于卷积结构的ResNet、DenseNet分类模型具有更高的分类准确率,可达96%,同时测试集的混淆矩阵表明Vision Transformer分类模型在样本不均衡数据集中对于小样本类别也具有较高的准确率,具有实际部署、推理的应用价值。

An artificial neural network for detecting weld position in arc welding process

A kind of self organizing artificial neural network used for weld detection is presented in this paper, and its concepts and issues are discussed. The network can transform the weld visual information into typical patterns and match with the weld data collected on line, and so realize the accurate detection of the weld position in arc welding process.

Performance Comparison of Vision Transformer- and CNN-Based Image Classification Using Cross Entropy: A Preliminary Application to Lung Cancer Discrimination from CT Images

This study evaluates the performance and reliability of a vision transformer (ViT) compared to convolutional neural networks (CNNs) using the ResNet50 model in classifying lung cancer from CT images into four categories: lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), large cell carcinoma (LULC), and normal. Although CNNs have made significant advancements in medical imaging, their limited capacity to capture long-range dependencies has led to the exploration of ViTs, which leverage self-attention mechanisms for a more comprehensive global understanding of images. The study utilized a dataset of 748 lung CT images to train both models with standardized input sizes, assessing their performance through conventional metrics—accuracy, precision, recall, F1 score, specificity, and AUC—as well as cross entropy, a novel metric for evaluating prediction uncertainty. Both models achieved similar accuracy rates (95%), with ViT demonstrating a slight edge over ResNet50 in precision and F1 scores for specific classes. However, ResNet50 exhibited higher recall for LULC, indicating fewer missed cases. Cross entropy analysis showed that the ViT model had lower average uncertainty, particularly in the LUAD, Normal, and LUSC classes, compared to ResNet50. This finding suggests that ViT predictions are generally more reliable, though ResNet50 performed better for LULC. The study underscores that accuracy alone is insufficient for model comparison, as cross entropy offers deeper insights into the reliability and confidence of model predictions. The results highlight the importance of incorporating cross entropy alongside traditional metrics for a more comprehensive evaluation of deep learning models in medical image classification, providing a nuanced understanding of their performance and reliability. While the ViT outperformed the CNN-based ResNet50 in lung cancer classification based on cross-entropy values, the performance differences were minor and may not hold clinical significance. Therefore, it may be premature to consider replacing CNNs with ViTs in this specific application.

Dynamic Coordination of Uncalibrated Hand/Eye Robotic System Based on Neural Network

A nonlinear visual mapping model is presented to replace the image Jacobian relation for uncalibrated hand/eye coordination. A new visual tracking controller based on artificial neural network is designed. Simulation results show that this method can drive the static tracking error to zero quickly and keep good robustness and adaptability at the same time. In addition, the algorithm is very easy to be implemented with low computational complexity.

基于Vision Transformer的高速公路监控场景天气识别

针对高速公路场景的天气识别问题,提出了基于Vision Transformer的天气种类自动识别算法。采集了广东省多条高速公路的监控视频图像,建立了高速公路监控场景下天气分类数据库,包括晴天、雨天、浓雾等天气场景。为了分析Vision Transformer自注意力机制的特性,利用特征图可视化手段对全连接层特征图的分类贡献作分析,分类贡献图像显示Vision Transformer分类网络的分类判断依据与人类视觉有相似之处,能有效关注到相同监控场景下不同天气图像的差异性。试验过程中,通过迁移学习方法以及数据增强的图片预处理方法提高了分类网络性能与鲁棒性,并测试对比了Vision Transformer与传统卷积神经网络在天气分类任务中的分类精度和识别速度。试验结果表明:由于具有自注意力机制,相比传统卷积神经网络,Vision Transformer特征提取能力更为高效,在天气图像分类任务中最高识别精度达到99.1%;在中小型数据集的分类任务中,压缩网络自注意力结构能保证识别精度的同时显著提高分类速度,达到与轻量化卷积神经网络相同的分类速度。试验过程中压缩自注意力层的Vision Transformer网络与原网络相比,识别精度仅降低了0.1%,识别速度提高了15.9%。

基于改进的Vision Transformer杂草生长周期识别方法研究

通过对农业杂草生长周期的识别分类,达到精准除草的目的。通过使用多卷积模块,使得改进的Vision Transformer模型提取到更加丰富的特征和语义信息,实现对杂草图像生长周期的识别分类。实验结果表明,运用改进Vision Transformer网络模型比传统卷积神经网络识别准确率提升4%。

RECOGNITION OF WEAR PARTICLES IN LUBRICATING OIL USING LVQ NEURAL CLASSIFIER

A technique for wear particle identification using computer vision system is described. The computer vision system employs LVQ Neural Networks as classifier to recognize the surface texture of wear particles in lubricating oil and determine the conditions of machines. The recognition process includes four stages:(1)capturing image from ferrographies containing wear particles;(2) digitising the image and extracting features;(3) learning the training data selected from the feature data set;(4) identifying the wear particles and generating the result report of machine condition classification. To verify the technique proposed here, the recognition results of several typical classes of wear particles generated at the sliding and rolling surfaces in a diesel engine are presented.

Learning-based force servoing control of a robot with vision in an unknown environment

A learning-based control approach is presented for force servoing of a robot with vision in an unknown environment. Firstly, mapping relationships between image features of the servoing object and the joint angles of the robot are derived and learned by a neural network. Secondly, a learning controller based on the neural network is designed for the robot to trace the object. Thirdly, a discrete time impedance control law is obtained for the force servoing of the robot, the on-line learning algorithms for three neural networks are developed to adjust the impedance parameters of the robot in the unknown environment. Lastly, wiping experiments are carried out by using a 6 DOF industrial robot with a CCD camera and a force/torque sensor in its end effector, and the experimental results confirm the effecti veness of the approach.

基于Vision Transformer的中文唇语识别

唇语识别作为一种将唇读视频转换为文本的多模态任务,旨在理解说话者在无声情况下表达的意思.目前唇语识别主要利用卷积神经网络提取唇部视觉特征,捕获短距离像素关系,难以区分相似发音字符的唇形.为了捕获视频图像中唇部区域像素之间的长距离关系,文中提出基于Vision Transformer(ViT)的端到端中文句子级唇语识别模型,融合ViT和门控循环单元(Gate Recurrent Unit,GRU),提高对嘴唇视频的视觉时空特征提取能力.具体地,首先使用ViT的自注意力模块提取嘴唇图像的全局空间特征,再通过GRU对帧序列时序建模,最后使用基于注意力机制的级联序列到序列模型实现对拼音和汉字语句的预测.在中文唇语识别数据集CMLR上的实验表明,文中模型的汉字错误率较低.

Vision Transformer的瞳孔定位方法

为了解决现有瞳孔定位方法易受瞳孔图像质量的约束,采用CNN提取图像的局部特征,通过Transformer的编码器获得全局依赖关系,发掘出更为准确的瞳孔中心信息,在公开数据集上对比了主流的DeepEye和VCF瞳孔定位模型。结果表明:提出的基于混合结构的Vision Transformer瞳孔定位方法在5像素误差内瞳孔中心的检测率比DeepEye提升了30%,比VCF提升了20%。

Selection for high quality pepper seeds by machine vision and classifiers

This research aimed to improve selection of pepper seeds for separating high-quality seeds from low-quality seeds. Past research has shown that seed vigor is significantly related to the seed color and size, thus several physical features were identified as candidate predictors of high seed quality. Image recognition software was used to automate recognition of seed feature quality using 400 kernels of pepper cultivar 101. In addition, binary logistic regression and a neural network were applied to determine models with high predictive value of seed germination. Single-kernel germination tests were conducted to validate the predictive value of the identified features. The best predictors of seed vigor were determined by the highest correlation observed between the physical features and the subsequent fresh weight of seedlings that germinated from the 400 seeds. Correlation analysis showed that fresh weight was significantly positively correlated with eight physical features： three color features （R, a＊, brightness）, width, length, projected area, and single-kernel density, and weight. In contrast, fresh weight significantly negatively correlated with the feature of hue. In analyses of two of the highest correlating single features,＇ germination percentage increased from 59.3 to 71.8% when a＊〉3, and selection rate peaked at 57.8%. Germination percentage increased from 59.3 to 79.4%, and the selection rate reached 76.8%, when single-kernel weight 〉0.0064 g. The most effective model was based on a multilayer perceptron （MLP） neural network, consisting of 15 physical traits as variables, and a stability calculated as 99.4%. Germination percentage in a calibration set of seeds was 79.1% and the selection rate was 90.0%. These results indicated that the model was effective in predicting seed germination based on physical features and could be used as a guide for quality control in seed selection. Automated systems based on machine vision and model classifiers can contribute to reducing the costs and labor required in the selection of pepper seeds.