结合改进CBAM和MobileNetV2算法对小麦病斑粒分类

[目的]小麦病斑粒智能检测对于高效、快速、准确评估麦粒的品级有重要意义。现有小麦病斑粒分类的深度神经网络存在参数数量大、运算复杂等缺点,不便于在移动端部署模型,影响了小麦病斑粒现场分类的效率。本文提出了一种用于小麦病斑粒分类的轻量级神经网络算法。[方法]本研究在轻量化网络MobileNetV2的基础上进行改进,加入改进的CBAM(convolutional block attention module)注意力机制,对改进的模型全整型量化后在边缘计算设备上部署。利用该模型对4种小麦籽粒(赤霉病粒、腥黑穗病粒、破损粒和正常粒)分类。[结果]相比于改进前的MobileNetV2网络,结合改进的注意力机制和MobileNetV2网络在准确率、精准率、召回率上分别提升3.15%、3%和3%;全整型量化后的改进模型对小麦赤霉病粒、腥黑穗病粒、破损粒和正常粒的识别准确率分别达到99%、94%、99%和96%。该模型大小仅有2.36 MB,在边缘计算设备的单次推理时间仅为96.95 ms。[结论]本文改进后的算法模型的准确率提升、大小减少、推理速度加快,可为小麦病斑粒分类模型去冗余提供指导。

关联成像技术在光学信息加密方面的研究现状及展望

光学信息安全技术因其高速、多维、并行等特点,成为现代加密技术的主要研究内容。而关联成像技术作为一种基于光场高阶关联获取物体信息的主动成像机制,具有结构简单、灵敏性好、抗干扰能力强等特点,同时其数据压缩降维的特性为光学信息安全技术提供了新的思路。基于此,根据应用场景的不同,详细介绍了关联成像技术在图像加密、保密传输、信息隐藏和认证及密钥分发等方面的国内外发展历程和应用现状,分析了关联成像加密方案存在的安全漏洞,讨论实际应用中尚存在的问题,并对关联成像加密方案的未来发展趋势进行了展望。

Detector development at the Back-n white neutron source

Purpose Back-n is a white neutron beamline at China spallation neutron source,which was established in the year of 2018.It is a powerful facility for nuclear data measurement,neutron detector calibration,and radiation effect research.Method A series of detectors were built for different experiments,including beam monitoring,beam profile measurement,neutron induced secondaries(fission fragments,light charged particles and gamma)cross section measurement,and neutron resonance radiography,etc.A common digitization electronics and a cluster-based DAQ were developed for these detector systems.Most detectors have been employed at Back-n and serviced for experiments from the beginning of the beamline running.Results and conclusion As an overview of detectors of Back-n,the details of the detector design and the experiment performing are described in this paper.Some developing systems,e.g.,MTPC and B-MCP,are also included.