基于K-means聚类和RF算法的葡萄霜霉病检测分级方法

针对自然环境复杂背景下葡萄霜霉病检测分级困难的问题,提出了一种基于语义分割结合K-means聚类和随机森林算法的葡萄霜霉病检测分级方法,实现对葡萄霜霉病快速分级。构建了葡萄霜霉病数据集,采用HRNet v2+OCR网络建立葡萄叶片语义分割模型,提取复杂环境下葡萄叶片;采用K-means聚类算法将葡萄叶片分解为若干子区域图像,并标记少量数据集进行随机森林算法学习,实现葡萄叶片病斑分割与提取;同时在叶片提取和病斑提取过程中,设计一种像素尺寸变换方法,解决图像分辨率引起的精度低问题。基于HRNet v2+OCR网络的葡萄叶片分割模型的准确率为98.45%,平均交并比为97.23%;融合K-means聚类和随机森林(RF)算法的葡萄叶片正面、反面和正反面霜霉病病害分级准确率分别为52.59%、73.08%和63.32%,病害等级误差小于等于2级时的病害分级准确率分别为88.67%、96.97%和92.98%。研究结果表明,基于K-means聚类和随机森林算法的葡萄霜霉病检测分级方法能够准确地分割自然环境复杂背景下的葡萄叶片和葡萄霜霉病病斑,并实现葡萄霜霉病分级,为葡萄霜霉病精准防治提供了方法和...     

基于改进YOLO v4的群体棉种双面破损检测方法

针对研究人员难以利用计算机视觉对棉种这类尺寸较小的物体进行双面检测,导致检测效果不佳的问题,设计了一款新型棉种检测分选装置,利用亚克力板在强光和白色背景下透明的特点,将棉种通过上料装置滑入透明亚克力板的凹槽中,随着转盘的转动,同一批棉种的正反两面图像分别由2个不同位置的CCD相机采集得到.利用改进YOLO v4的目标检...     

基于改进YOLO v4的玉米种子外观品质检测方法

针对玉米种子在外观品质检测中需要快速识别与定位的需求,提出了一种基于改进YOLO v4的目标检测模型,同时结合四通道(RGB+NIR)多光谱图像,对玉米种子外观品质进行了识别与分类。为了减少改进后模型的参数量,本文将主干特征提取网络替换为轻量级网络MobileNet V1。为了进一步提升模型的性能,通过试验研究了空间金字塔池化(Spatial pyramid pooling, SPP)结构在不同位置上对模型性能的影响,最终选取改进YOLO v4-MobileNet V1模型对玉米种子外观品质进行检测。试验结果表明,模型的综合评价指标平均F1值和mAP达到93.09%和98.02%,平均每检测1幅图像耗时1.85 s,平均每检测1粒玉米种子耗时0.088 s,模型参数量压缩为原始模型的20%。四通道多光谱图像的光谱波段可扩展到可见光范围之外,并能够提取出更具有代表性的特征信息,并且改进后的模型具有鲁棒性强、实时性好、轻量化的优点,为实现种子的高通量质量检测和优选分级提供了参考。     

融合特征金字塔与可变形卷积的高密度群养猪计数方法

针对猪只人工计数方法消耗时间和劳动力,育肥猪较为活跃且喜好聚集,图像中存在大量的高密度区域,导致猪只之间互相粘连、遮挡等问题,基于SOLO v2实例分割算法,提出了一种自然养殖场景下融合多尺度特征金字塔与二代可变形卷积的高密度群养猪计数模型。通过优化模型结构来减少计算资源的消耗与占用。将科大讯飞给出的猪只计数的公开数据集划分为猪只分割数据集和猪只盘点测试集,利用猪只分割数据集获得较好的分割模型,然后在猪只盘点测试集中测试盘点准确率,实现猪群分割和猪只计数。实验结果表明,本文提出的高密度猪只计数模型的分割准确率达到96.7%,且模型内存占用量为256 MB,为改进前的2/3,实现了遮挡、粘连和重叠情况下的猪只个体高准确率分割。在含有500幅猪只图像计数测试集中,模型计算猪只数量误差为0时的图像数量为207幅,较改进前提高26%。模型计算猪只数量误差小于2头猪的图像数量占测试图像总数量的97.2%。模型计算猪只数量误差大于3头猪的图像数量占总体图像数量比例仅为1%。最后,对比基于YOLO v5的群养猪计数方法,本文模型具有更优的分割效果和计数准确率,验证了本文方法对群养猪只计数的有效性。因...     

基于YOLO v5-TL的褐菇采摘视觉识别-测量-定位技术

为实现褐菇高效、精准、快速的自动化采摘,针对工厂化褐菇的种植特点,提出一种基于YOLO v5迁移学习(YOLO v5-TL)结合褐菇三维边缘信息直径动态估测法的褐菇原位识别-测量-定位一体化方法。首先,基于YOLO v5-TL算法实现复杂菌丝背景下的褐菇快速识别;再针对锚框区域褐菇图像进行图像增强、去噪、自适应二值化、形态学处理、轮廓拟合进行褐菇边缘定位,并提取边缘点和褐菇中心点的像素坐标;最后基于褐菇三维边缘信息的直径动态估测法实现褐菇尺寸的精确测量和中心点定位。试验结果表明单帧图像平均处理时间为50ms,光照强度低、中、高情况下采摘对象识别平均成功率为91.67%,其中高光强时识别率达100%,菇盖的尺寸测量平均精度为97.28%。研究表明,本文提出的YOLO v5-TL结合褐菇三维边缘信息直径动态估测法可实现工厂化种植环境下褐菇识别、测量、定位一体化,满足机器人褐菇自动化采摘需求。     

基于改进YOLO v3的自然场景下冬枣果实识别方法

为实现自然场景下冬枣果实的快速、精准识别,考虑到光线变化、枝叶遮挡、果实密集重叠等复杂因素,基于YOLO v3深度卷积神经网络提出了一种基于改进YOLO v3（YOLO v3-SE）的冬枣果实识别方法。YOLO v3-SE模型利用SE Net 的SE Block结构将特征层的特征权重校准为特征权值,强化了有效特征,弱化了低效或无效特征,提高了特征图的表现能力,从而提高了模型识别精度。YOLO v3-SE模型经过训练和比较,选取0.55作为置信度最优阈值用于冬枣果实检测,检测结果准确率P为88.71%、召回率R为83.80%、综合评价指标F为86.19%、平均检测精度为82.01%,与YOLO v3模型相比,F提升了2.38个百分点,mAP提升了4.78个百分点,检测速度无明显差异。为检验改进模型在冬枣园自然场景下的适应性,在光线不足、密集遮挡和冬枣不同成熟期的情况下对冬枣果实图像进行检测,并与YOLO v3模型的检测效果进行对比,结果表明,本文模型召回率提升了2.43~5.08个百分点,F提升了1.75~2.77个百分点,mAP提升了2.38~4.81个百分点,从而验证了本文模型的有效性。     

基于改进YOLO v4模型的马铃薯中土块石块检测方法

为实现收获后含杂马铃薯中土块石块的快速检测和剔除,提出了一种基于改进YOLO v4模型的马铃薯中土块石块检测方法。YOLO v4模型以CSPDarknet53为主干特征提取网络,在保证检测准确率的前提下,利用通道剪枝算法对模型进行剪枝处理,以简化模型结构、降低运算量。采用Mosaic数据增强方法扩充图像数据集(8621幅图像),对模型进行微调,实现了马铃薯中土块石块的检测。测试表明,剪枝后模型总参数量减少了94.37%,模型存储空间下降了187.35 MB,前向运算时间缩短了0.02 s,平均精度均值(Mean average precision, mAP)下降了2.1个百分点,说明剪枝处理可提升模型性能。为验证模型的有效性,将本文模型与5种深度学习算法进行比较,结果表明,本文算法mAP为96.42%,比Faster R-CNN、Tiny-YOLO v2、YOLO v3、SSD分别提高了11.2、11.5、5.65、10.78个百分点,比YOLO v4算法降低了0.04个百分点,模型存储空间为20.75 MB,检测速度为78.49 f/s,满足实际生产需要。     

Use of a cyclic high-pressure sodium lamp to inhibit flowering of chrysanthemum and velvet sage

Photoperiod is commonly controlled in the commercial production of ornamental crops to induce or prevent flowering. Flower induction in short-day (SD) plants can be prevented or delayed when the natural daylength is short by providing low-intensity lighting during the dark period. A stationary high-pressure sodium (HPS) lamp with an oscillating aluminum parabolic reflector (cyclic HPS) has been developed to provide intermittent lighting to greenhouse crops. We determined the efficacy of a cyclic HPS lamp at preventing flowering in SD plants garden chrysanthemum [Chrysanthemum × grandiflorum (Ramat.) Kitam.] ‘Bianca’, pot chrysanthemum ‘Auburn’, and velvet sage (Salvia leucantha L.) relative to traditional night interruption (NI) lighting strategies. Plants were grown in a glass-glazed greenhouse at a mean daily temperature of 19.5–20.7 °C with natural SD photoperiods. NI lighting was delivered during the middle of the night (2230–0230 h) from a 600 W cyclic HPS lamp mounted at one gable end of the greenhouse or from incandescent (INC) lamps that were illuminated for the entire 4 h (CONT INC) or for 6 min every 30 min for 4 h. Plants under cyclic HPS were grown at lateral distances of 1, 4, 7, 10, or 13 m from under the lamp. Control plants were grown under an uninterrupted 15 h skotoperiod. As the distance from the cyclic HPS lamp increased from 1 to 13 m, the maximum irradiance measured during the NI decreased from 25.4 to 0.3 μmol m⁻² s⁻¹ and time to visible inflorescence (VI) and the number of nodes at VI decreased. All species had a VI within 54 d, but ≤10% of plants flowered when grown at a lateral distance of 1 or 4 m from the cyclic HPS lamp or under CONT INC. Plants grown without NI had a VI 2 to 15 d earlier and flowered 7 to 24 d earlier than plants grown at 10 or 13 m from the cyclic HPS. All garden chrysanthemums flowered under cyclic INC, whereas velvet sage and pot chrysanthemum had 15% and 35% flowering, respectively. These results indicate that a cyclic HPS lamp can be used effectively to delay flower induction and prevent flowering in these species when NI is delivered at ≥2.4 μmol m⁻² s⁻¹.     

施氮水平对7S亚基缺失大豆根系形态和结瘤固氮的影响

为有效推广功能型大豆7S亚基缺失品种,以7S亚基缺失大豆品系东富2号为研究对象,设置4种施氮水平(纯N),N0(0 mg·kg~(-1))、N1(25 mg·kg~(-1))、N2(50 mg·kg~(-1))、N3(75 mg·kg~(-1)),采用桶栽法研究大豆根系形态和结瘤固氮对不同施氮水平的响应。结果表明:N1(25 mg·kg~(-1))水平下根系干重加大,根冠比增大,根瘤固氮潜力高,单株产量较高。N2(50 mg·kg~(-1))水平下根长、根表面积、根体积在生育后期增长较快,根系干重较大,根冠比低,固氮酶活性最高,单株籽粒产量最高。N3(75 mg·kg~(-1))水平下植株干重较大,无效生长较多,根瘤数少,固氮潜力和根冠比低,单株产量不高。综合籽粒产量和根系特性指标,功能型大豆7S亚基缺失品系东富2号的适宜施肥量为25~50 mg·kg~(-1)。     

Agronomic traits of soybean cultivars released in different decades after grafting record‐yield cultivar as rootstock

Breeders have seldom considered the selection for root traits during the genetic improvement in soybean. It is hypothesized that grain yield would be increased by the root function improvement, especially for the current cultivars. The objective of this grafting experiment was to determine the effect of record‐yield cultivars L14 or Z35 as rootstocks on agronomic traits of cultivars released in different decades. A total of 11 cultivars, released in different decades, were used to graft onto L14 or Z35 rootstocks. The agronomic traits were measured in the pot‐culture experiments. Grafting cultivars released in different decades onto L14 or Z35 rootstocks resulted in higher yield, 100‐seed mass and apparent harvest index as compared with those of non‐grafted or self‐grafted plants. Grain yield gain of cultivars grafted onto record‐yield cultivar rootstocks was 0.40 g/plant/year from 1966 to 2006, which was larger than that of non‐grafts and self‐grafts (0.27 g/plant/year). The yield of current cultivars should increase more if their root functions were improved.