Application of UAV-Based Imaging and Deep Learning in Assessment of Rice Blast Resistance

doi:10.1016/j.rsci.2023.06.005

Abstract

Abstract:

Rice blast is regarded as one of the major diseases of rice. Screening rice genotypes with high resistance to rice blast is a key strategy for ensuring global food security. Unmanned aerial vehicles (UAV)-based imaging, coupled with deep learning, can acquire high-throughput imagery related to rice blast infection. In this study, we developed a segmented detection model (called RiceblastSegMask) for rice blast detection and resistance evaluation. The feasibility of different backbones and target detection models was further investigated. RiceblastSegMask is a two-stage instance segmentation model, comprising an image-denoising backbone network, a feature pyramid, a trinomial tree fine-grained feature extraction combination network, and an image pixel codec module. The results showed that the model combining the image-denoising and fine-grained feature extraction based on the Swin Transformer and the feature pixel matching feature labels with the trinomial tree recursive algorithm performed the best. The overall accuracy for instance segmentation of RiceblastSegMask reached 97.56%, and it demonstrated a satisfactory accuracy of 90.29% for grading unique resistance to rice blast. These results indicated that low-altitude remote sensing using UAV, in conjunction with the proposed RiceblastSegMask model, can efficiently calculate the extent of rice blast infection, offering a new phenotypic tool for evaluating rice blast resistance on a field scale in rice breeding programs.

Key words: rice blast, segmentation detection, trinomial tree, Swin Transformer, unmanned aerial vehicle

Lin Shaodan, Yao Yue, Li Jiayi, Li Xiaobin, Ma Jie, Weng Haiyong, Cheng Zuxin, Ye Dapeng. Application of UAV-Based Imaging and Deep Learning in Assessment of Rice Blast Resistance[J]. Rice Science, 2023, 30(6): 652-660.

Figures/Tables 7

References 30

[1]	Alvarez J, Martinez E, Diezma B. 2021. Application of hyperspectral imaging in the assessment of drought and salt stress in magneto-primed Triticale seeds. Plants, 10(5): 835.
[2]	An G Q, Xing M F, He B B, Kang H Q, Shang J L, Liao C H, Huang X D, Zhang H G. 2021. Extraction of areas of rice false smut infection using UAV hyperspectral data. Remote Sens, 13(16): 3185.
[3]	Anderson R S. 1991. The origins of the International Rice Research Institute. Minerva, 29(1): 61-89.
[4]	Araus J L, Cairns J E. 2014. Field high-throughput phenotyping: The new crop breeding frontier. Trends Plant Sci, 19(1): 52-61.
[5]	Ariza-Sentís M, Valente J, Kooistra L, Kramer H, Mücher S. 2023. Estimation of spinach (Spinacia oleracea) seed yield with 2D UAV data and deep learning. Smart Agric Technol, 3: 100129.
[6]	Atkins J. 1967. An international set of rice varieties for differentiating races of Piricularia oryzae. Phytopathology, 57: 297-301.
[7]	Bal S S, Kumar S, Kar A, Mishra I, Panda P. 2019. Screening of mungbean germplasm for resistance against important diseases under rice fallow condition. J Plant Prot Environ, 16(1): 34-37.
[8]	Bolya D, Zhou C, Xiao F Y, Lee Y J. 2020. YOLACT: Real-time instance segmentation. 2019 IEEE/CVF International Conference on Computer Vision (ICCV). October 27-November 2, 2019. Seoul, Korea (South): IEEE: 9156-9165.
[9]	Brugger A, Schramowski P, Paulus S, Steiner U, Kersting K, Mahlein A K. 2021. Spectral signatures in the UV range can be combined with secondary plant metabolites by deep learning to characterize barley-powdery mildew interaction. Plant Pathol, 70(7): 1572-1582.
[10]	Cai N, Zhou X G, Yang Y B, Wang J, Zhang D Y, Hu R J. 2018. Use of UAV images to assess narrow brown leaf spot severity in rice. Int J Precis Agric Aviat, 1(1): 38-42.
[11]	Cen H Y, Wan L, Zhu J P, Li Y J, Li X R, Zhu Y M, Weng H Y, Wu W K, Yin W X, Xu C, Bao Y D, Feng L, Shou J Y, He Y. 2019. Dynamic monitoring of biomass of rice under different nitrogen treatments using a lightweight UAV with dual image-frame snapshot cameras. Plant Methods, 15(1): 1-16.
[12]	Feng S, Zhao D X, Guan Q, Li J P, Liu Z Y, Jin Z Y, Li G M, Xu T Y. 2022. A deep convolutional neural network-based wavelength selection method for spectral characteristics of rice blast disease. Comput Electron Agric, 199: 107199.
[13]	Feng X P, Zhan Y H, Wang Q, Yang X F, Yu C L, Wang H Y, Tang Z Y, Jiang D A, Peng C, He Y. 2020. Hyperspectral imaging combined with machine learning as a tool to obtain high- throughput plant salt-stress phenotyping. Plant J, 101(6): 1448-1461.
[14]	Jin P F, Wang Y, Tan Z, Liu W B, Miao W G. 2020. Antibacterial activity and rice-induced resistance, mediated by C₁₅surfactin A, in controlling rice disease caused by Xanthomonas oryzae pv. oryzae. Pestic Biochem Physiol, 169: 104669.
[15]	Ke L, Danelljan M, Li X, Tai Y W, Tang C K, Yu F. 2022. Mask transfiner for high-quality instance segmentation. 2022 IEEE/ CVF Conference on Computer Vision and Pattern Recognition (CVPR). June 18-24, 2022. New Orleans, LA, USA: IEEE: 4402-4411.
[16]	Kim Y, Roh J H, Kim H. 2017. Early forecasting of rice blast disease using long short-term memory recurrent neural networks. Sustainability, 10(2): 34.
[17]	Laha G S, Singh R, Ladhalakshmi D, Sunder S, Prasad M S, Dagar C S, Babu V R. 2017. Importance and management of rice diseases: A global perspective. In: Chauhan B S, Jabran K, Mahajan G. Rice Production Worldwide. Cham, Switzerland: Springer International Publishing: 303-360.
[18]	Li D L, Bai D, Tian Y, Li Y H, Zhao C S, Wang Q, Guo S Y, Gu Y Z, Luan X Y, Wang R Z, Yang J L, Hawkesford M J, Schnable J C, Jin X L, Qiu L J. 2023. Time series canopy phenotyping enables the identification of genetic variants controlling dynamic phenotypes in soybean. J Integr Plant Biol, 65(1): 117-132.
[19]	Li W, Feng X S, Zha K W, Li S N, Zhu H S. 2021. Summary of target detection algorithms. J Phys: Conf Ser, 1757(1): 012003.
[20]	Li Y, Qian M Y, Liu P F, Cai Q, Li X Y, Guo J W, Yan H, Yu F Y, Yuan K, Yu J, Qin L M, Liu H X, Wu W, Xiao P Y, Zhou Z W. 2019. The recognition of rice images by UAV based on capsule network. Cluster Comput, 22(4): 9515-9524.
[21]	Liang J Y, Cao J Z, Sun G L, Zhang K, van Gool L, Timofte R. 2021. SwinIR: Image restoration using Swin Transformer. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops. 11-17 October, 2021. Montreal, BC, Canada: IEEE: 1833-1844.
[22]	Lin S D, Li X B, Yang B Y, Chen C, He W C, Weng H Y, Ye D P. 2022. Detecting citrus Huanglongbing from few-shot microscopic images using an improved DETR. Trans Chin Soc Agric Eng, 38(12): 216-223. (in Chinese with English abstract)
[23]	Ma X Q, Duan G H, Chen H F, Tang P, Su S Y, Wei Z X, Yang J. 2022. Characterization of infected process and primary mechanism in rice Acuce defense against rice blast fungus, Magnaporthe oryzae. Plant Mol Biol, 110(3): 219-234.
[24]	Nugent P W, Shaw J A, Jha P, Scherrer B, Donelick A, Kumar V. 2018. Discrimination of herbicide-resistant kochia with hyper- spectral imaging. J Appl Remote Sens, 12(1): 016037.
[25]	Oerke E C, Leucker M, Steiner U. 2019. Sensory assessment of Cercospora beticola sporulation for phenotyping the partial disease resistance of sugar beet genotypes. Plant Methods, 15(1): 1-12.
[26]	Shahriar S A, All Imtiaz A, Hossain M B, Husna A, Eaty M N K. 2020. Review: Rice blast disease. Annu Res Rev Biol, 35(1): 50-64.
[27]	Wan L, Li Y J, Cen H Y, Zhu J P, Yin W X, Wu W K, Zhu H Y, Sun D W, Zhou W J, He Y. 2018. Combining UAV-based vegetation indices and image classification to estimate flower number in oilseed rape. Remote Sens, 10(9): 1484.
[28]	Zhang D Y, Zhou X G, Zhang J, Lan Y B, Xu C, Liang D. 2018. Detection of rice sheath blight using an unmanned aerial system with high-resolution color and multispectral imaging. PLoS One, 13(5): e0187470.
[29]	Zhang J N, Feng X P, Wu Q G, Yang G F, Tao M Z, Yang Y, He Y. 2022. Rice bacterial blight resistant cultivar selection based on visible/near-infrared spectrum and deep learning. Plant Methods, 18(1): 1-16.
[30]	Zhou X, Zheng H B, Xu X Q, He J Y, Ge X K, Yao X, Cheng T, Zhu Y, Cao W X, Tian Y C. 2017. Predicting grain yield in rice using multi-temporal vegetation indices from UAV-based multi- spectral and digital imagery. ISPRS J Photogramm Remote Sens, 130: 246-255.

Backbone	Training time (h)	Box loss	Mask loss	Classification loss	Segmentation loss
ResNet50	60.88	0.024	0.024	0.069	0.022
ResNet101	36.22	0.026	0.026	0.083	0.022
Swin-Tiny	30.23	0.097	0.065	0.040	0.044
Swin-Base	38.27	0.067	0.044	0.027	0.034

Backbone	Training time (h)	Box loss	Mask loss	Classification loss	Segmentation loss
ResNet50	60.88	0.024	0.024	0.069	0.022
ResNet101	36.22	0.026	0.026	0.083	0.022
Swin-Tiny	30.23	0.097	0.065	0.040	0.044
Swin-Base	38.27	0.067	0.044	0.027	0.034

Disease level	ResNet50	ResNet101	Swin-Tiny	Swin-Base	ResNet50	ResNet101	Swin-Tiny	Swin-Base
Box average precision					Mask average precision
1	100.00^a	100.00^a	80.00	100.00^a	90.00	97.33^a	60.00	80.00
2	98.88 ^a	98.86	98.01	98.77	95.95^a	95.42	90.52	93.07
3	97.88	98.16	98.21^a	97.97	94.65	95.46^a	91.13	93.96
4	98.99	99.00^a	98.29	98.98	97.27^a	96.73	91.45	93.98
5	100.00^a	100.00^a	100.00^a	100.00^a	100.00^a	95.05	95.05	95.05
6	98.99	99.00^a	98.42	99.00^a	97.04^a	96.80	91.47	93.91
7	98.80^a	98.77	97.90	98.74	96.26^a	95.59	90.17	92.95
8	97.97^a	97.85	97.29	97.86	93.99^a	93.71	89.31	91.08
9	98.87	98.89	98.25	98.98^a	97.55	97.33	92.76	97.56^a

Disease level	ResNet50	ResNet101	Swin-Tiny	Swin-Base	ResNet50	ResNet101	Swin-Tiny	Swin-Base
Box average precision					Mask average precision
1	100.00^a	100.00^a	80.00	100.00^a	90.00	97.33^a	60.00	80.00
2	98.88 ^a	98.86	98.01	98.77	95.95^a	95.42	90.52	93.07
3	97.88	98.16	98.21^a	97.97	94.65	95.46^a	91.13	93.96
4	98.99	99.00^a	98.29	98.98	97.27^a	96.73	91.45	93.98
5	100.00^a	100.00^a	100.00^a	100.00^a	100.00^a	95.05	95.05	95.05
6	98.99	99.00^a	98.42	99.00^a	97.04^a	96.80	91.47	93.91
7	98.80^a	98.77	97.90	98.74	96.26^a	95.59	90.17	92.95
8	97.97^a	97.85	97.29	97.86	93.99^a	93.71	89.31	91.08
9	98.87	98.89	98.25	98.98^a	97.55	97.33	92.76	97.56^a

Model	AP (IOU 0.50)	AP (IOU 0.75)	AP (IOU 0.90)	mAP (IOU 0.50‒0.95)
YOLACT	92.09	90.50	48.78	77.12
YOLACT++	93.10	90.53	49.86	77.83
Fast R-CNN	87.02	82.03	33.56	67.54
Mask R-CNN	95.03	87.14	57.35	79.84
Transfiner	95.08	85.33	58.33	79.58
RiceblastSegMask	96.83	87.33	56.33	80.16