基于深度学习的番茄叶部病害识别模型

番茄叶病种类多、成因复杂,其预防和识别难度较大。传统基于机器学习的方法多靠人工识别,需要一定的专家经验,且具有主观性强、识别准确率不高等缺点。为实现番茄叶病特征的自动提取,并提高识别准确率,提出一种基于深度学习的番茄叶病识别模型。该模型基于卷积神经网络对番茄叶部病害特征进行自动提取,获得高维特征后,采用PCA降维算法去除冗余特征;从增大类间距离并减小类内距离的角度改进了softmax分类器,提高了识别准确率。将该模型在CrowdAI提供的数据集上进行了仿真验证,结果表明,该模型能够对番茄叶部常见10种病害进行自动提取特征和识别,综合识别准确率达到95%以上。     

Steganography using a 3-player game

Image steganography aims to securely embed secret information into cover images. Until now, adaptive embedding algorithms such as S-UNIWARD or Mi-POD, were among the most secure and most often used methods for image steganography. With the arrival of deep learning and more specifically, Generative Adversarial Networks (GAN), new steganography techniques have appeared. Among them is the 3-player game approach, where three networks compete against each other. In this paper, we propose three different architectures based on the 3-player game. The first architecture is proposed as a rigorous alternative to two recent publications. The second takes into account stego noise power. Finally, our third architecture enriches the second one with a better interaction between embedding and extracting networks. Our method achieves better results compared to existing works Hayes and Danezis (2017), Zhu et al. (2018), and paves the way for future research on this topic.     

Suppression of forced vibration due to chip segmentation in ultrasonic elliptical vibration cutting of titanium alloy Ti–6Al–4V

Ultrasonic elliptical vibration cutting of titanium alloy Ti–6Al–4V is investigated in this research. Because products made of Ti–6Al–4V alloy are usually designed for possessing low-rigidity structures or good-quality cut surfaces, machining requirements such as low cutting forces and slow rate of tool wear need to be fulfilled for realization of their precision machining. Therefore, the ultrasonic elliptical vibration cutting is applied as a novel machining method for those products. Machinability of Ti–6Al–4V alloy by the ultrasonic elliptical vibration cutting with cemented carbide tools is examined to figure out suitable cutting conditions for precision machining of Ti–6Al–4V alloy. As experimental results, generated chips, cutting forces, and profiles of cut surfaces are indicated. A forced vibration problem occurred due to the segmented chip formation, which is also well-known in the ordinary non-vibration cutting. Therefore, characteristics of the forced vibration due to the chip segmentation are investigated in this research. Through the experiments, it is found that the frequency and magnitude of the forced vibration have relation with the average uncut chip thickness and cutting width. Especially, it is found that the averaging effect can suppress the forced vibration, i.e. the chip segmentation tends to occur randomly over the large cutting width, and hence the force fluctuations with random phases tend to cancel each other as the cutting width increases relatively against the average uncut chip thickness. Based on the investigations, a new practical strategy to suppress the forced vibration due to chip segmentation is proposed and verified. Using the proposed method significantly decreased cutting forces and good quality of surfaces are obtained when the forced vibration is suppressed compared to the ordinary non-vibration cutting results. Therefore, the results suggest that the precision machining can be realized without sacrificing the machining efficiency by increasing the width of cut and decreasing the average uncut chip thickness.     

基于Faster R-CNN的颜色导向火焰检测

基于深度特征的目标检测方法Faster R-CNN在火焰检测任务上存在检测效率低的问题,因此提出了基于颜色引导的抛锚策略。该策略设计火焰颜色模型来限制锚的生成,即利用火焰颜色约束锚的生成区域,从而减少了初始锚的数量,提升了计算效率。为了进一步提高网络的计算效率,将区域生成网络中的卷积层替换成掩膜卷积。为了验证所提方法的检测效果,采用BoWFire和Corsician数据集进行验证。实验结果表明,该方法实际检测速度相较于原Faster R-CNN提高了10.1%,BoWFire上该方法的火焰检测F值为0.87,Corsician上该方法的准确度可达99.33%。所提方法可以提高火焰检测的效率,并能够准确检测图像中的火焰。     

重新找回人工智能的可解释性

针对深度神经网络AI研究的可解释性瓶颈，指出刚性逻辑（数理形式逻辑）和二值神经元等价，二值神经网络可转换成逻辑表达式，有强可解释性。深度神经网络一味增加中间层数来拟合大数据，没有适时通过抽象把最小粒度的数据（原子）变成粒度较大的知识（分子），再把较小粒度的知识变成较大粒度的知识，把原有的强可解释性淹没在中间层次的汪洋大海中。要支持多粒度的知识处理，需把刚性逻辑扩张为柔性命题逻辑（命题级数理辩证逻辑），把二值神经元扩张为柔性神经元，才能保持强可解释性。本文详细介绍了从刚性逻辑到柔性逻辑的扩张过程和成果，最后介绍了它们在AI研究中的应用，这是重新找回AI研究强可解释性的最佳途径。     

基于STM32单片机的大学生体能监测仪设计与实现

体能监控仪作为生命体征监测设备,对于辅助大学生完成体能训练有很大帮助;基于STM32单片机技术,设计开发集成了多种传感器的一种大学生体能监测装备,它能够有效的检测包括心率检测、人体的步数、运动圈速、消耗卡路里等信息;该设计采用了STM32作为主控芯片,采用模块化的设计思路;在系统硬件设计方面,设计了包括STM32控制系统、心率获取电路、三轴传感器电路、显示电路以及按键电路等;同时将软件模块拆分成了初始化子程序、心率获取子程序,步数获取子程序,显示子程序等;Proteus仿真结果显示,使用STM32单片机技术设计的体能监测系统,能够运行稳定,测量心率范围60次/min~180次/min,可在心率异常时报警,同时检测步数的准确率达98%,满足体能监测任务需要,实验结果符合预期;为STM32单片机在体育运动研究信息化中应用提供参考。     

Measurement of vertical and horizontal vibrations of a probe for acoustic evaluation of food texture

Two‐dimensional (vertical and horizontal) vibrations of a wedge‐type probe upon food rupture were evaluated separately using two accelerometers placed perpendicular to a guide rod of a swing‐arm device for texture evaluation of the flesh of three varieties of apples and three types of potato chips. Voltage signals from the accelerometers were filtered using a half‐octave band‐pass filter. The energy texture index (ETI), based on kinetic energy of the vertical or horizontal probe vibrations, was calculated over low to high frequency bands (no. 1–21). The spectra for the flesh of the three varieties of apples included two common peaks for vertical ETI at band no. 11 (1,120–1,600 Hz) and 19 (17,920–25,600 Hz) and horizontal ETI at band no. 1 (0–10 Hz) and 15 (4,480–6,400 Hz). The spectra for the three types of potato chips had a common ETI peak at band no. 11 (1,120–1,600 Hz) for horizontal ETI and at no. 15 (4,480–6,400 Hz) for vertical ETI. The three apple varieties gave rise to different intensities of vertical and horizontal ETIs while the two peaks were maintained. The thick potato chip type had higher vertical and horizontal ETIs than the thin and soft varieties in most bands; however, the thin type had the highest vertical ETIs only in lower bands (0–200 Hz). The soft type had the lowest vertical and horizontal ETI. The above results suggest that different food textures can be distinguished by two‐dimensional vibration analyses of probe insertion into a food sample based on frequency bands.     

智能温度测量系统设计

以AT89C52单片机作为微控制器，设计了一个智能温度测量系统。该系统通过DS18B20温度传感器实现环境温度的采集和A/D转换，模数转化后的电信号送入AT89C52单片机，再通过单片机将其送入LCD1602数码管进行显示。通过实验结果可以看出，相比较传统的温度采集方法，该系统能够快速准确地检测温度，并且检测精度可以达到0.1 ；同时，该系统具有过温点设定值调节及温度上下限蜂鸣提示的优点。     

基于多尺度双线性卷积神经网络的多角度下车型精细识别

针对多角度下车辆出现一定的尺度变化和形变导致很难被准确识别的问题，提出基于多尺度双线性卷积神经网络（MS-B-CNN）的车型精细识别模型。首先，对双线性卷积神经网络（B-CNN）算法进行改进，提出MS-B-CNN算法对不同卷积层的特征进行了多尺度融合，以提高特征表达能力；此外，还采用基于中心损失函数与Softmax损失函数联合学习的策略，在Softmax损失函数基础上分别对训练集每个类别在特征空间维护一个类中心，在训练过程中新增加样本时，网络会约束样本的分类中心距离，以提高多角度情况下的车型识别的能力。实验结果显示，该车型识别模型在CompCars数据集上的正确率达到了93.63%，验证了模型在多角度情况下的准确性和鲁棒性。