Detection method of high temperature stress of tomato at seedling stage based on thermal infrared and RGB images
-
摘要:目的
针对实际生产场景中番茄苗期生长遇到的高温胁迫问题，提出一种基于热红外和RGB图像的番茄苗期高温胁迫检测方法。
方法首先，通过番茄苗期热红外图像反演获取番茄冠层温度参数，采用偏最小二乘(Partial least squares, PLS)模型提取冠层温度特征指标；然后，建立采用3种不同主干特征提取网络的Mask-RCNN模型，通过迁移学习的方式将番茄苗期RGB图像输入Mask-RCNN模型，进行高温胁迫症状实例分割，得到番茄苗期胁迫症状特征指标；最后，利用提取的温度和胁迫症状特征指标构建分级数据集，输入高温胁迫分级模型，得到高温胁迫等级。
结果基于PLS模型提取的冠层温度特征指标累计贡献率达95.45%；基于ResNet101+Mask-RCNN的高温胁迫症状分割网络对番茄苗期轻度和重度胁迫的分割精度最高，均值平均查准率(Mean average precision, mAP)分别为77.3%和73.8%；基于温度和胁迫症状特征指标构建的4种高温胁迫分级模型中，反向传播神经网络(Back propagation neural network, BPNN)获得最好的高温胁迫分级效果，分级准确率达95.6%。
结论该方法对番茄苗期高温胁迫检测效果较好，可为番茄苗期高温胁迫早期精准检测和快速自动预警提供技术支撑。
Abstract:ObjectiveAiming at the problem of high temperature stress encountered in the growth of tomato at seedling stage in actual production scenarios, a method for detecting high temperature stress of tomato at seedling stage based on thermal infrared and RGB images was proposed.
MethodFirstly, the tomato canopy temperature parameters were obtained by inversion through the thermal infrared image of tomato plant at seedling stage, and the canopy temperature characteristic indicators were extracted by the partial least squares (PLS) model. Then, a Mask-RCNN model using three different backbone feature extraction networks was established, and the RGB images of tomato seedlings were input into the Mask-RCNN model by means of transfer learning for instance segmentation of high temperature stress symptoms. The characteristic indicators of tomato stress symptoms at seedling stage were obtained. Finally, the extracted temperature and stress symptom characteristic indicators were used to construct a hierarchical data set and fed into the high temperature stress classification model to obtain the high temperature stress level.
ResultThe cumulative contribution rate of the canopy temperature characteristic indicators extracted based on the PLS model reached 95.45%. The high temperature stress symptom segmentation network based on ResNet101+Mask-RCNN had the highest segmentation accuracy for mild and severe stress of tomato at seedling stage, with mean average precision (mAP) of 77.3% and 73.8% respectively. Among the four high temperature stress grading models constructed based on temperature and stress symptom characteristic indicators, the back propagation neural network (BPNN) showed the best high temperature stress grading performance with the grading accuracy rate of 95.6%.
ConclusionThe proposed method in this study achieves better detection performance for high temperature stress of tomato at seedling stage, and provides a technical support for early detection and rapid automatic warning of high temperature stress of tomato at seedling stage.
-
Keywords:
- Seedling stage /
- Tomato /
- High temperature /
- Thermal infrared image /
- Mask-RCNN /
- Stress detection
-
普通大蓟马Megalurothrips usitatus又名豆大蓟马、豆花蓟马，隶属于缨翅目蓟马科大蓟马属，主要分布于澳大利亚、马来西亚、斯里兰卡、菲律宾、斐济、印度、日本等[1-3]，在我国海南、台湾、广东、广西、湖北、贵州、陕西等地也均有发生为害[4-5]。据报道，该虫有28种寄主，其中16种为豆科植物，目前它已成为危害华南地区豆科作物的主要害虫[6-9]，田间调查和室内试验均表明豇豆为其嗜好寄主[10-11]。普通大蓟马主要以锉吸式口器取食豇豆幼嫩组织的汁液，可造成叶片皱缩、生长点萎缩、豆荚痂疤等，严重影响豇豆品质[12-13]。此外，该虫体积小、发生量大、隐秘性强，大部分时间都躲在花中取食，从豇豆苗期至采收期均可为害[14-15]，以上特点均增加了农户的防治难度。当其为害严重时，农户只能增加施药频率和施药量，这也导致该虫对多种常用化学农药产生了严重的抗药性[16-17]。
目前关于普通大蓟马的研究主要集中在生物学特性[18]及综合防治技术[19-20]等层面，随着抗药性的不断发展与研究的不断深入，从分子层面解析普通大蓟马的抗药性机制和寄主选择机制等以寻求新型绿色防控方法势在必行，室内种群的大规模饲养是展开这些研究的基础。化蛹基质作为影响昆虫种群规模的关键因子，韩云等[21]曾指出普通大蓟马在含水量(w)为15%的砂壤土中羽化率显著高于砂土、壤土和黏土，但不适用于室内大规模饲养，因为实际应用中，存在土壤类型无法明确区分、配制砂壤土会增加人工饲养的工作量等问题。土壤以外的其他基质对普通大蓟马化蛹的适合度鲜见研究报道。
本研究以普通大蓟马为试验对象，室内观测其在沙子、蛭石和厨房用纸3种基质及无基质条件下的羽化规律，分析该虫对不同化蛹基质的适合度，以期为普通大蓟马的室内大规模饲养提供基础资料，为该虫的综合治理提供理论依据。
1. 材料与方法
1.1 供试材料
普通大蓟马于2017年采自广东省广州市增城区朱村豇豆田，采回后在RXZ-500C型智能人工气候箱(宁波江南仪器厂)内用豇豆豆荚饲养，饲养条件为温度(26±6) ℃，光照周期12 h光∶12 h暗，相对湿度(70±5)%。室内饲养多代后，选取发育一致的老熟2龄若虫(以体色变为橙红色为标准)进行室内试验。
供试基质包括沙子、蛭石、锯末和厨房用纸，并以无基质作为空白对照。试验前将沙子、蛭石和锯末置于DHG-9140型电热恒温鼓风干燥箱(上海精宏实验设备有限公司)中105 ℃恒温烘烤6 h备用。
1.2 试验方法
首先称取过筛烘干后的沙子50 g 3组，分别加入2.5、3.5和4.5 mL蒸馏水，充分混匀，配制成含水量(w)分别为5%、7%和9%的沙子化蛹基质；称取过筛烘干后的蛭石10 g 3组，分别加入10.0、12.5和15.0 mL蒸馏水，充分混匀，配制成含水量(w)分别为20%、25%和30%的蛭石化蛹基质；称取过筛烘干后的蛭石10 g 3组，分别加入12.5、15.0和17.5 mL蒸馏水，充分混匀，配制成含水量(w)分别为25%、30%和35%的锯末化蛹基质。将以上基质分别转移至350 mL玻璃组培瓶内，基质深度均为5 cm，将厨房用纸对折成合适大小后平铺在组培瓶底部作为基质。在所有基质上放置纱网，再加入1根新鲜的豇豆豆荚(长度约4~5 cm)，分别接入50头普通大蓟马老熟2龄若虫，用250目纱布封口后置于人工气候箱中饲养，每日观察并记录成虫羽化数量。每个处理设6次重复。设置不加入任何化蛹基质的空白对照。
含水量的测定方法按以下公式[22]进行：
含水量=实际含水质量/烘干后基质质量×100%。
1.3 数据分析
运用SPSS 24.0软件进行试验数据处理分析，不同基质及含水量对普通大蓟马羽化率、蛹历期和性比(雄性∶雌性)的影响采用单因素方差分析，并运用Duncan’s法检验差异显著性。
2. 结果与分析
2.1 不同基质对普通大蓟马羽化率、蛹历期和性比的影响
普通大蓟马在不同基质中的羽化率、蛹历期和性比具有显著差异(图1)。由图1A可知，普通大蓟马在厨房用纸中的羽化率显著高于其他基质，为54.33%，其次为含水量5%(w)的沙子，羽化率为44.67%；锯末最不适宜于普通大蓟马羽化，在含水量(w)为25%、30%、35%的锯末中普通大蓟马的羽化率分别为10.33%、5.33%、16.67%，显著低于空白对照与其他基质。
![]()
图 1 不同基质对普通大蓟马羽化率、发育历期和性比(雄性∶雌性)的影响1~3分别为含水量(w)为5%、7%和1%的沙子，4~6分别为含水量(w)为20%、25%和30%的蛭石，7~9分别为含水量(w)为25%、30%和35%锯末，10：厨房用纸，11：无基质；各图中的不同小写字母表示差异显著(P<0.05，Duncan’s法)Figure 1. Effects of different substrates on eclosion rate, pupa developmental period and male-female ratio of Megalurothrips usitatus1: Sand with 5% moisture, 2: Sand with 7% moisture, 3: Sand with 10% moisture, 4: Vermiculite with 20% moisture, 5: Vermiculite with 25% moisture, 6: Vermiculite with 30% moisture, 7: Sawdust with 25% moisture, 8: Sawdust with 30% moisture, 9: Sawdust with 35% moisture, 10: Kitchen paper, 11: No substrate; Different lowercase leters in the same figure indicated significant difference among different substrate (P<0.05, Duncan’s method)由图1B可知，普通大蓟马在含水量5%(w)的沙子中蛹的发育历期最短，为5.29 d，其次为含水量7%(w)的沙子，为6.01 d，在其他基质中的蛹期则无显著差异，在6.14~7.16 d。
由图1C可知，普通大蓟马在含水量30%(w)的蛭石中性比最高，为0.60，含水量10%(w)的沙子和30%(w)的蛭石性比相对较低，分别为0.12和0.06，在其他基质中性比无显著差异。
2.2 不同基质条件下普通大蓟马的羽化情况
由表1数据可知，沙子含水量(w)为5%时普通大蓟马羽化最早，始于第2天；其次为蛭石，羽化始于第4天，其他条件下羽化均始于第3天；以锯末为基质时羽化最晚，始于第5天。沙子含水量(w)为5%和厨房用纸条件下，羽化高峰出现在第5天，羽化率分别为21%和22.67%；次高峰在第6天，羽化率分别为14.33%和21%。沙子含水量(w)为9%、锯末以及空白对照下羽化高峰出现在第7天，其他条件下羽化高峰均出现在第6天。不同基质类型及含水量条件下，普通大蓟马的羽化均结束于第8天或第9天，与不同基质培养条件下普通大蓟马蛹期之间的差异相对应。
表 1 不同基质对普通大蓟马逐日羽化率的影响1)Table 1. Effects of differents substrates on daily eclosion rate of Megalurothrips usitatus% t/d 沙子含水量(w) Water content in sand 蛭石含水量(w) Water content in vermiculite 5% 7% 9% 20% 25% 30% 1 0 0 0 0 0 0 2 1.67±0.42c 0 0 0 0 0 3 1.00±1.68c 0 0 0 0 0 4 1.33±0.67c 5.33±0.33c 0.33±0.33b 0 0 0 5 21.00±3.82a 5.33±2.17b 2.67±1.91b 3.00±2.30bc 10.33±3.48ab 0.33±0.33b 6 14.33±4.66b 17.33±1.76a 2.67±1.91b 11.67±2.09a 14.67±3.33a 7.67±2.22a 7 2.33±0.80c 5.00±0.85b 8.67±1.84a 6.33±2.28b 7.67±1.74bc 6.67±1.52a 8 0.67±0.42c 0.67±0.67c 0.67±0.42b 4.00±1.35bc 4.00±1.37cd 1.67±0.94b 9 0 0 0.33±0.33b 0.67±0.42b 0 0.67±0.42b 10 0 0 0 0 0 0 3. 讨论与结论
化蛹基质的类型对普通大蓟马化蛹具有一定影响，本研究发现锯末和蛭石不适宜于普通大蓟马化蛹，锯末和蛭石不同含水量条件下大蓟马的羽化率都显著低于空白对照。有研究指出土壤中砂土含量低于30%时，蓟马若虫不能化蛹[23]，蓟马在砂壤土中的羽化率也显著高于砂土、黏土、壤土等单一土壤[21]。
化蛹基质的含水量对普通大蓟马化蛹具有显著影响，本研究发现当沙子含水量(w)为5%时，羽化率仅次于厨房用纸，高达44.67%，与孟国玲等[23]关于豆带蓟马Taenithripsglycines在含水量(w)为5.7%时羽化率最高(43.63%)的报道相对一致。韩云等[21]研究发现普通大蓟马在含水量(w)为15%的砂壤土中羽化率最高，为52.08%，而土壤含水量(w)5%时羽化率仅为6.67%。这与本研究结果不符，究其原因可能是不同类型的基质吸水力与保水力不同，导致在相同的绝对含水量下湿度有差异。此外，有研究曾指出高含水量不利于蓟马化蛹[24]，这与本研究结果相一致，沙子含水量(w)5%时的羽化率显著高于含水量(w)7%和10%。
在本研究中，成虫性比普遍低于1∶1，含水量(w)30%的蛭石羽化性比最高，为0.6，含水量(w)30%锯末最低，为0.06，其他处理的性比无显著差异，为0.12~0.48。张念台[8]和谭柯[24]在田间调查的结果也显示其成虫性比低于1∶1，后代总是偏于雌性，谭柯[24]则表示后代偏雌性可能是蓟马暴发的原因之一。这与本研究结果相一致，后代偏于雌性。
本研究发现普通大蓟马在厨房用纸中的羽化率最高，蛹发育历期与其他基质相比无明显差异，且以厨房用纸为化蛹基质时，可以清楚地观察到普通大蓟马蛹期的形态特征变化，可以随时根据试验需求收集不同时期的若虫或成虫。虽然沙子含水量(w)5%时蛹发育历期最短且羽化率也较高，但蓟马一旦入土化蛹便无法继续观察形态或收集虫体。因此，本试验条件下，厨房用纸是最适合室内普通大蓟马大量饲养的化蛹基质。
-
![]()
图 1 试验整体流程图
PCA：主成分分析；PLS：偏最小二乘；NB：朴素贝叶斯；SVM：支持向量机；kNN：k−最近邻；BPNN：反向传播神经网络
Figure 1. Overall flow chart of the test
PCA: Principal component analysis; PLS: Partial least squares; NB: Naive Bayesian; SVM: Support vector machine; kNN: k-nearest neighbor; BPNN: Back propagation neural network
![]()
图 2 番茄幼苗热红外图像温度分布直方图
Figure 2. Histogram of temperature distribution of thermal infrared image of tomato seedling
![]()
图 3 番茄幼苗冠层能量分布三维曲面图
Figure 3. 3D surface map of energy distribution in canopy of tomato seedlings
![]()
图 4 不同高温胁迫症状的番茄幼苗叶片RGB图像
Figure 4. RGB images of tomato seedling leaves with different symptoms of high temperature stress
![]()
图 5 Mask-RCNN架构
FPN：特征金字塔网络；RPN：区域生成网络；RoI：感兴趣区域
Figure 5. Mask-RCNN architecture
FPN: Feature pyramid network; RPN: Region proposal network; RoI: Region of interest
![]()
图 7 3种不同主干特征提取网络不同程度胁迫预测结果
Figure 7. Prediction results of different stresses on three different backbone feature extraction networks
![]()
图 8 Mask-RCNN+ResNet101模型损失函数值变化曲线
Figure 8. Change curve of loss function value of Mask-RCNN+ResNet101 model
![]()
图 9 4种不同模型测试集分类结果混淆矩阵
绿色方格代表分类正确样本，其中数字为样本数量和所占总体样本的比例；红色方格代表分类错误样本；浅灰色方格中数字代表测试集样本每一行和列的真阳性率(绿色)和假阴性率(红色)；深灰色方格中数字代表测试集样本的总体准确率(绿色)和总体错误率(红色)
Figure 9. Confusion matrices of classification results of test sets of four different models
Green squares represent correctly classified samples, where the numbers are the number of samples and the proportion of the total sample; Red squares represent misclassified samples; The numbers in the light gray squares represent the true positive rate (green) and false negative rate (red) for each row and column of the test set sample; The numbers in the dark gray squares represent the overall accuracy (green) and overall error (red) of the test set samples
表 1 番茄幼苗冠层热红外温度参数变化表
Table 1 Variation table of thermal infrared temperature parameters of tomato seedling canopy
℃ 试验时刻1)Experiment time 最大温度 Maximum temperature 最小温度 Minimum temperature 最大温差 Maximum temperature difference 平均温度 Average temperature 温度标准差 Temperature standard deviation D1 12:00 30.3 21.7 8.6 24.6230 1.5155 D1 16:00 31.6 21.7 9.9 25.1910 1.8274 D2 12:00 32.4 22.9 9.5 25.1721 1.3691 D2 16:00 31.2 22.7 8.5 25.9238 1.4817 D3 12:00 31.1 23.1 8.0 26.8236 1.7791 D3 16:00 32.0 23.1 8.9 27.3416 1.7418 D4 12:00 31.8 24.3 7.5 27.0165 1.3433 D4 16:00 32.9 24.1 8.8 27.5235 1.5168 D5 12:00 33.1 24.6 8.5 29.0355 1.7989 D5 16:00 34.2 25.2 9.0 30.0699 1.8027 D6 12:00 33.5 24.8 8.7 29.7350 1.7264 D6 16:00 34.4 25.2 9.2 30.1302 1.8432 1) D1~D6分别表示第1~6天 1) D1-D6 indicate day 1-6, respectively 
下载: 导出CSV
表 2 主干特征提取网络性能参数
Table 2 Performance parameters of backbone feature extraction network
网络 Network 深度 Depth 大小/MB Size 参数量 Millions parameter quantity 图像输 入大小 Image input size 特征提取层 Feature extraction layer 池化层输出大小 Output size of pooling layer ResNet50 50 96 25.6 224×224 Block_13_expand_relu 14×14 ResNet101 101 167 44.6 224×224 Mixed7 17×17 MobileNet 54 13 3.5 224×224 Res4b_22_relu 14×14
下载: 导出CSV
表 3 温度特征指标的相关系数和主成分贡献率
Table 3 Correlation coefficient and principal component contribution rate of temperature characteristic indicators
特征指标 Characteristic indicator 相关系数 Correlation coefficient 主成分贡献率/% Principal component contribution rate 最大值( $ {T}_{1} $) Maximum 0.1451 3.24 最小值( $ {T}_{2} $) Minimum −0.0385 0.07 平均值( $ {T}_{3} $) Average −0.1328 0.23 标准差( $ {T}_{4} $) Standard deviation −0.0437 0.01 规范标准化值( $ {T}_{5} $) Canonical normalized −0.1466 4.17 规范替换值( $ {T}_{6} $) Canonical replacement −0.1657 6.44 变异系数( $ {T}_{7} $) Coefficient of variation 0.2749 34.37 冠气温差( $ {T}_{8} $) Crown temperature difference 0.1563 2.28 信息熵( $ {T}_{9} $) Information entropy 0.2134 16.62 20~23 ℃温度频率( $ {T}_{10} $) Temperature frequency −0.1360 0.25 23~26 ℃温度频率( $ {T}_{11} $) Temperature frequency −0.2041 9.76 26~29 ℃温度频率( $ {T}_{12} $) Temperature frequency 0.0826 0.03 29~32 ℃温度频率( $ {T}_{13} $) Temperature frequency 0.2279 21.85 32~35 ℃温度频率( $ {T}_{14} $) Temperature frequency 0.1491 0.68
下载: 导出CSV
表 4
${\rm{IoU}}=0.5$ 时3种不同主干特征提取网络Mask-RCNN模型分割精度Table 4 Segmentation accuracy of Mask-RCNN model of three different backbone feature extraction networks when IoU=0.5
主干特征提取网络 Backbone feature extraction network 胁迫等级 Image stress level 查准率/% Precision 查全率/% Recall $ {F}_{1} $/% $ \mathrm{m}\mathrm{A}\mathrm{P} $/% ResNet50 中度 Mild 90.4 50.5 64.8 75.7 重度 Severe 83.2 45.9 59.2 71.6 ResNet101 中度 Mild 91.3 52.4 66.6 77.3 重度 Severe 85.7 47.2 60.9 73.8 MobileNet 中度 Mild 86.0 46.8 60.6 69.9 重度 Severe 80.5 43.4 56.4 66.5
下载: 导出CSV
表 5 不同分类网络模型的苗期番茄高温胁迫分级精度
Table 5 Classification accuracy of high temperature stress of tomato at seedling stage with different classification network model
模型类别 Model category 准确率/% Accuracy 查准率/% Precision 查全率/% Recall $ {F}_{1} $/% ${{\rm{MCC}}}$/% 朴素贝叶斯 Naive Bayesian (NB) 90.6 92.7 89.9 91.3 81.0 支持向量机 Support vector machine (SVM) 93.3 96.8 91.0 93.8 86.8 k−最近邻 k-nearest neighbor (kNN) 91.7 92.7 91.8 92.2 83.3 BP神经网络 BP neural network (BPNN) 95.6 96.9 94.9 95.9 91.1
下载: 导出CSV
-
[1] ZHANG J, JIANG X D, LI T L, et al. Photosynthesis and ultrastructure of photosynthetic apparatus in tomato leaves under elevated temperature[J]. Photosynthetica, 2014, 52(3): 430-436. doi: 10.1007/s11099-014-0051-8
[2] 巩文睿, 金萍, 钟启文. 设施农业物联网技术应用现状与发展建议[J]. 农业科技管理, 2017, 36(4): 20-23. doi: 10.16849/j.cnki.issn1001-8611.2017.04.006 [3] ZHOU R, KONG L, WU Z, et al. Physiological response of tomatoes at drought, heat and their combination followed by recovery[J]. Physiologia Plantarum, 2019, 165(2): 144-154.
[4] WEN J Q, JIANG F L, LIU M, et al. Identification and expression analysis of Cathepsin B-like protease 2 genes in tomato at abiotic stresses especially at high temperature[J]. Scientia Horticulturae, 2021, 277(10): 109799.
[5] TORRES G M, LOLLATO R P, OCHSNER T E. Comparison of drought probability assessments based on atmospheric water deficit and soil water deficit[J]. Agronomy Journal, 2013, 105(2): 428-436. doi: 10.2134/agronj2012.0295
[6] WISHART J, GEORGE T S, BROWN L K, et al. Field phenotyping of potato to assess root and shoot characteristics associated with drought tolerance[J]. Plant and Soil, 2014, 378(1/2): 351-363.
[7] ZHOU Z, MAJEED Y, DIVERRES NARANJO G, et al. Assessment for crop water stress with infrared thermal imagery in precision agriculture: A review and future prospects for deep learning applications[J]. Computers and Electronics in Agriculture, 2021, 182: 106019.
[8] CHANDEL A K, KHOT L R, YU L X. Alfalfa (Medicago sativa L. ) crop vigor and yield characterization using high-resolution aerial multispectral and thermal infrared imaging technique[J]. Computers and Electronics in Agriculture, 2021, 182: 105999.
[9] HOU M J, TIAN F, ORTEGA-FARIAS S, et al. Estimation of crop transpiration and its scale effect based on ground and UAV thermal infrared remote sensing images[J]. European Journal of Agronomy, 2021, 131: 126389. doi: 10.1016/j.eja.2021.126389
[10] DAS S, CHAPMAN S, CHRISTOPHER J, et al. UAV-thermal imaging: A technological breakthrough for monitoring and quantifying crop abiotic stress to help sustain productivity on sodic soils: A case review on wheat[J]. Remote Sensing Applications Society and Environment, 2021, 23: 100583. doi: 10.1016/j.rsase.2021.100583
[11] ZHU W, CHEN H, CIECHANOWSKA I, et al. Application of infrared thermal imaging for the rapid diagnosis of crop disease[J]. IFAC-PapersOnLine, 2018, 51(17): 424-430. doi: 10.1016/j.ifacol.2018.08.184
[12] LACERDA L N, SNIDER J L, COHEN Y, et al. Using UAV-based thermal imagery to detect crop water status variability in cotton[J]. Smart Agricultural Technology, 2022, 2: 100029. doi: 10.1016/j.atech.2021.100029
[13] TIAN F, HOU M, QIU Y, et al. Salinity stress effects on transpiration and plant growth under different salinity soil levels based on thermal infrared remote (TIR) technique[J]. Geoderma, 2020, 357: 113961. doi: 10.1016/j.geoderma.2019.113961
[14] MA S, BI Y, ZHANG Y, et al. Thermal infrared imaging study of water status and growth of arbuscular mycorrhizal soybean (Glycine max) under drought stress[J]. South African Journal of Botany, 2022, 146: 58-65. doi: 10.1016/j.sajb.2021.09.037
[15] BANERJEE K, KRISHNAN P. Normalized sunlit shaded index (NSSI) for characterizing the moisture stress in wheat crop using classified thermal and visible images[J]. Ecological Indicators, 2020, 110: 47-59.
[16] SINGH A, GANAPATHYSUBRAMANIAN B, SINGH A K, et al. Machine learning for high-throughput stress phenotyping in plants[J]. Trends in Plant Science, 2016, 21(2): 110-124. doi: 10.1016/j.tplants.2015.10.015
[17] SINGH A K, SINGH, GANAPATHYSUBRAMANIAN B, SARKAR S, et al. Deep learning for plant stress phenotyping: Trends and future perspectives[J]. Trends in Plant Science, 2018, 23(10): 883-898. doi: 10.1016/j.tplants.2018.07.004
[18] ANAMI B S, MALVADE N N, PALAIAH S. Deep learning approach for recognition and classification of yield affecting paddy crop stresses using field images[J]. Artificial Intelligence in Agriculture, 2020, 4: 12-20. doi: 10.1016/j.aiia.2020.03.001
[19] ZHOU J, LI J X, WANG C S, et al. Crop disease identification and interpretation method based on multimodal deep learning[J]. Computers and Electronics in Agriculture, 2021, 189: 106408.
[20] MOON S K, AMJAD M, QURESHI M, et al. Use of deep learning techniques for identification of plant leaf stresses: A review[J]. Sustainable Computing: Informatics and Systems, 2020, 28: 100443. doi: 10.1016/j.suscom.2020.100443
[21] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL]. arXiv: 1409. 1556. (2014-01-01) [2022-01-10]. https://arXiv.org/abs/1409.1556.
[22] SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the inception architecture for computer vision[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA: IEEE. 2016: 2818-2826.
[23] WU S J, ZHONG S H, LIU Y. Deep residual learning for image steganalysis[J]. Multimedia Tools and Applications, 2018, 77(9): 10437-10453. doi: 10.1007/s11042-017-4440-4
[24] JI M M, WU Z B. Automatic detection and severity analysis of grape black measles disease based on deep learning and fuzzy logic[J]. Computers and Electronics in Agriculture, 2022, 193: 106718. doi: 10.1016/j.compag.2022.106718
[25] LIPPMANN R, BABBEN S, MENGER A, et al. Development of wild and cultivated plants under global warming conditions[J]. Current Biology, 2019, 29(24): R1326-R1338. doi: 10.1016/j.cub.2019.10.016
[26] 易平涛, 李伟伟, 郭亚军. 线性无量纲化方法的结构稳定性分析[J]. 系统管理学报, 2014, 23(1): 104-110. doi: 10.3969/j.issn.1005-2542.2014.01.015 [27] MA J Y, MA Y, LI C. Infrared and visible image fusion methods and applications: A survey[J]. Information Fusion, 2019, 45: 153-178.
[28] WESTERHUIS J A, KOURTI T, MACGREGOR J F. Analysis of multiblock and hierarchical PCA and PLS models[J]. Journal of Chemometrics, 1998, 12(5): 301-321.
[29] HE K M, GKIOXARI G, DOLLÁR P, et al. Mask-R-CNN[C]// 2017 IEEE International Conference on Computer Vision. Veniu, Italy: IEEE, 2017: 2961-2969.
[30] REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
[31] LONG J, SHELHAMER E, DARRELL T. Fully convolutional networks for semantic segmentation[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston, USA: IEEE, 2015: 640-651.
[32] HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, USA: IEEE, 2015: 1512-1516.
[33] RUSSELL B C, TORRALBA A, MURPHY K P, et al. LabelMe: A database and web-based tool for image annotation[J]. International Journal of Computer Vision, 2008, 77(1/2/3): 157-173. doi: 10.1007/s11263-007-0090-8
[34] 穆龙涛, 高宗斌, 崔永杰, 等. 基于改进AlexNet的广域复杂环境下遮挡猕猴桃目标识别[J]. 农业机械学报, 2019, 50(10): 24-34. doi: 10.6041/j.issn.1000-1298.2019.10.003 [35] 乔虹, 冯全, 赵兵, 等. 基于Mask R-CNN的葡萄叶片实例分割[J]. 林业机械与木工设备, 2019, 47(10): 15-22. doi: 10.3969/j.issn.2095-2953.2019.10.003 [36] AMARO E G, CANALES J C, CABRERA J E, et al. Identification of diseases and pests in tomato plants through artificial vision[M]//Intelligent Computing Methodologies. Cham: Springer International Publishing, 2020: 98-109.
[37] LI Y R, YANG K M, GAO W, et al. A spectral characteristic analysis method for distinguishing heavy metal pollution in crops: VMD-PCA-SVM[J]. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 2021, 255: 119649. doi: 10.1016/j.saa.2021.119649
[38] ANAMI B S, MALVADE N N, PALAIAH S. Classification of yield affecting biotic and abiotic paddy crop stresses using field images[J]. Information Processing in Agriculture, 2020, 7(2): 272-285.
-
期刊类型引用(11)
1. 刘琅，李文秀，于凯波，吴鹍伦，周行，褚晶，吴朝晖. 控释肥与不同农药联合施用对水稻生长发育、产量和氮素利用率的影响. 江苏农业科学. 2025(02): 68-74 . 
百度学术
2. 何意林，沈彤，田天，李国利. 植物源农药5%香芹酚水剂的急性毒性初步研究. 毒理学杂志. 2024(01): 85-87 . 百度学术
3. 张月，宋明丹，塔林葛娃，李月梅. 有机无机肥配施对春小麦产量、养分吸收及土壤矿质氮残留的影响. 江苏农业科学. 2024(17): 80-88 . 百度学术
4. 张一帆，何瑞银，段庆飞，徐勇. 基于CFD-DEM的排肥用波纹管结构优化设计与试验. 浙江农业学报. 2023(01): 191-201 . 百度学术
5. 赵欢欢，付建涛，安玉兴，卢颖林，陈立君，孙东磊. 我国药肥研究现状及前景分析. 热带农业科学. 2023(02): 97-102 . 百度学术
6. 李文秀，吴鹍伦，刘琅，周行，褚晶，吴朝晖. 不同药肥处理对杂交早稻潭两优83生长发育及产量的影响. 杂交水稻. 2023(06): 127-134 . 百度学术
7. 仲凤翔，梅爱中，钱爱林，崔劲松，王春兰. 25%甲氧·茚虫威SC等药剂防治稻纵卷叶螟药效试验. 福建稻麦科技. 2022(01): 31-33 . 百度学术
8. 李文秀，周行，刘琅，吴朝晖. 稻作生产中水、肥、药高效利用及对水稻的影响研究进展. 河南农业科学. 2022(06): 1-12 . 百度学术
9. 邓家欣，韦继光，於虹，姜燕琴，曾其龙，刘红军，蒋佳峰. 不同施肥处理对高丛越橘幼苗生长和生理指标及土壤理化性质的影响. 植物资源与环境学报. 2021(02): 28-34 . 百度学术
10. 王辉. 水稻施肥中多种复合肥的肥效对比试验. 农业开发与装备. 2021(06): 153-154 . 百度学术
11. 于洋，侯新月，袁安丽，高月. 寒区水稻水肥管理技术研究进展. 水利科学与寒区工程. 2021(05): 78-81 . 百度学术
其他类型引用(4)
计量
- 文章访问数: 131
- HTML全文浏览量: 15
- PDF下载量: 316
- 被引次数: 15
