密集场景下群株生菜的有效分割与参数获取是植物工厂生长监测中的关键环节。针对群株生菜中个体生菜鲜质量提取问题,该研究提出一种利用实例分割模型提取个体生菜点云,再以深度学习点云算法预测个体鲜质量的方法。该方法以群株生菜为研究对象,利用深度相机采集群株生菜俯视点云,将预处理后的点云数据输入实例分割模型Mask3D中训练,实现背景与生菜个体的实例分割,之后使用鲜质量预测网络预测个体生菜鲜质量。试验结果表明,该模型实现了个体生菜点云的分割提取,无多检和漏检的情况。当交并比(intersection over union,IoU)阈值为0.75时,群株生菜点云实例分割的精确度为0.924,高于其他实例分割模型;鲜质量预测网络实现了直接通过深度学习处理点云数据,预测个体生菜鲜质量的目的,预测结果的决定系数R2值为0.90,均方根误差值为12.42 g,优于从点云中提取特征量,再回归预测鲜质量的传统方法。研究结果表明该研究预测生菜鲜质量的精度较高,为利用俯视单面点云提取群株生菜中个体生菜表型参数提供了一种思路。
为精准识别与分类不同花期杭白菊,满足自动化采摘要求,该研究提出一种基于改进YOLOv8s的杭白菊检测模型-YOLOv8s-RDL。首先,该研究将颈部网络(neck)的C2f(faster implementation of CSP bottleneck with 2 convolutions)模块替换为RCS-O...
详细信息
为精准识别与分类不同花期杭白菊,满足自动化采摘要求,该研究提出一种基于改进YOLOv8s的杭白菊检测模型-YOLOv8s-RDL。首先,该研究将颈部网络(neck)的C2f(faster implementation of CSP bottleneck with 2 convolutions)模块替换为RCS-OSA(one-shot aggregation of reparameterized convolution based on channel shuffle)模块,以提升骨干网络(backbone)特征融合效率;其次,将检测头更换为DyHead(dynamic head),并融合DCNv3(deformable convolutional networks v3),借助多头自注意力机制增强目标检测头的表达能力;最后,采用LAMP(layer-adaptive magnitude-based pruning)通道剪枝算法减少参数量,降低模型复杂度。试验结果表明,YOLOv8s-RDL模型在菊米和胎菊的花期分类中平均精度分别达到96.3%和97.7%,相较于YOLOv8s模型,分别提升了3.8和1.5个百分点,同时权重文件大小较YOLOv8s减小了6 MB。该研究引入TIDE(toolkit for identifying detection and segmentation errors)评估指标,结果显示,YOLOv8s-RDL模型分类错误和背景检测错误相较YOLOv8s模型分别降低0.55和1.26。该研究为杭白菊分花期自动化采摘提供了理论依据和技术支撑。
随着经济水平上升,汽车逐渐成为人们出行的主要选择之一。车内驾驶员的驾驶状态是影响行驶安全的重要因素,而当前的基于视觉识别的驾驶员动作和状态检测易受光照和遮挡等问题影响,且涉及用户隐私问题。毫米波雷达具有高探测精度,高集成、不受光线等因素影响、低成本等优点,已经广泛应用与体征信号、动作识别等领域,但目前对于驾驶姿态的动作识别种类较少。为此,本文基于77 GHz毫米波雷达,对驾驶员在车内动作进行信号采集,构建了包含静止、点头、左右环视、顿头(瞌睡)、前后剧烈晃动(急刹)、手部平移(抽烟)、手部抬起(打电话)七种动作的数据集。同时开发了基于VGG16-LSTM-CBAM的深度学习网络模型,对微多普勒频谱图进行分类识别。实验结果显示,本文提出的模型识别准确率达到99.16%,有效地提高了对驾驶员头手协同动作的识别精度。As the economic level rises, automobile gradually becomes one of the main choices for people’s traveling. The driving status of the driver in the car is an important factor that affects driving safety, and the current visual recognition-based driver action and status detection is susceptible to problems such as light and occlusion, and involves user privacy issues. Millimeter-wave radar has the advantages of high detection accuracy, high integration, insensitivity to light and other factors, and low cost, and thus has been widely used in the fields of body signals and action recognition. However, for recognition of driver’s postures, existing studies are limited to only a few actions. In this paper, using a 77 GHz millimeter wave radar, we constructed a dataset containing seven kinds of driver’s actions, including stationary, nodding head, left and right looking around, head-stopping (dozing), front and rear violent shaking (sharp braking), hand panning (smoking), hand lifting (phone call). A deep learning network model based on VGG16-LSTM-CBAM is also developed to classify and recognize micro-Doppler spectrograms. The experimental results show that the recognition accuracy of the proposed model reaches 99.16%, which effectively improves the recognition accuracy of driver actions.
暂无评论