University of Groningen
Computational intelligence & modeling of crop disease data in Africa
Owomugisha, Godliver
DOI:
10.33612/diss.130773079
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date: 2020
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Owomugisha, G. (2020). Computational intelligence & modeling of crop disease data in Africa. University of Groningen. https://doi.org/10.33612/diss.130773079
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
867. A low-cost 3-D printed smartphone add-on spectrometer for diagnosis of crop diseases inthe field
Table 7.2: Confusion matrix for Aspectra Mini with Extra trees
Healthy CBSD Healthy 56.4 43.6
CBSD 44.9 55.1
Table 7.3: Confusion matrix for Color Histograms with Extra trees
Healthy CBSD Healthy 59.5 40.5
CBSD 28.6 71.4
7.4 Discussion
We have presented an initial step towards the construction of an innovative low-cost spectrometer that can be used to diagnose disease in plants and infield. Our novel contribution in this area can be seen in the design of the prototype. Previous work (chapters 4 - 6) showed that spectrometry can gain the smallholder farmer an extra 8 weeks to apply an intervention before disease symptoms become visible. Our ex-periments in this study aimed at replacing the expensive spectrometer ($10K) with a cheap version to cost $5 ´ $8. While performance is clearly inferior to the one of the commercial spectrometer, we observe performance above mere guessing. This forms the basis for further improvements. One element to include in a future ver-sion is a powerful diffraction grating medium e.g. a prism. We will also experiment with different diodes. The explicit transformation of light emissions to actual spec-trograms should facilitate further improvements. Literature (Public Lab 2019) also shows efficacy for this type of handcrafted cheap DIY tools. We intend to leverage on that success to provide a diagnostic tool to be used by smallholder farmers in developing countries.
Chapter 8
Summary and Outlook
T
his thesis presents the application of machine learning techniques to solve a real world challenge related to pest and disease control in the agricultural sec-tor. The research investigated methods for early disease diagnosis with a novel ap-proach of identifying diseases before they became symptomatic and visible to the human eye.In Chapter 2, we provided the background on prototype based classification that has mostly been used in our studies. We found GMLVQ suitable to address most of our research questions due to its competitive or superior performance. The analysis of relevances provided additional insights and facilitated the identification of most important features.
In Chapter 3, we tackled the problem of crop disease detection using an image dataset captured with a mobile phone camera. As an initial step towards early dis-ease diagnosis, we investigated on disdis-ease incidence and severity measurements from cassava leaf images. We applied computer vision techniques to extract visual features of color and shape combined with classification techniques.
In Chapter 4, we considered disease diagnosis using spectral data. We compared two datasets: spectral data collected from leaves of the plant and leaf image data captured with a mobile phone camera. We analysed data from visibly affected parts of the leaf and parts that appear to be healthy, visibly. We analysed the obtained data by prototype based methods and standard classification models in a three-class clas-sification problem. Results point towards significant improvement in performance using spectral data and the possibility of early detection of disease before the crops become symptomatic, which for practical reasons is highly significant.
In Chapter 5, we answered several of our research questions. By nature, spec-tral data come with thousands of dimensions, therefore different wavelengths are analyzed in order to identify the most relevant spectral bands. To cope with the nominally high number of input dimensions of data, functional decomposition of the spectra is considered. The outlined classification task was addressed using GM-LVQ and compared with the standard classification techniques performed in the space of expansion coefficients.
cas-88 8. Summary and Outlook sava crop, once disease symptoms are visible in the aerial part of the plant, a lot of damage has been caused especially in the root of the plant which is majorly used as food. The novelty and usefulness of this research is evidenced in Chapter 6. The chapter presented results in using visible and near infrared spectral information to detect diseases in cassava crops before symptoms can be seen by the human eye. To test this hypothesis, we grew cassava plants in a screen house where they were inoculated with disease viruses and we monitored the plants over time collecting both spectral and plant tissue for wet chemistry analysis at each time step until the plants show disease. Our models in our case GMLVQ were able to detect cassava diseases one week after virus infection can be confirmed by wet lab chemistry, but several weeks before symptoms manifest on the plants.
Although the analysis of spectral data proved successful in our experiments, spectrometry devices are not commonly found on the market, ready to do auto-mated diagnosis of crop diseases. Also, most spectrometry devices require techni-cal knowledge to operate them, thus making them inapplicable for our end users (smallholder farmers). In our case, we purchased one expensive spectrometer (in 1000 USD) to achieve our research goals. In Chapter 7, we presented initial steps towards the development of a low-cost 3-D printed smartphone add-on spectrome-ter that can be used to diagnose crop diseases in the fields. The contribution of this chapter can be seen in the design of a diagnostic tool that is cheap and easy to use by smallholder farmers in developing countries in monitoring their crop status. The performance of the first prototype was not favourable, however, requirements for an improved version were stated.
8.1 Future work
The research presented in this thesis builds from previous studies as stated in the literature. The present work was transitioning from previous methods of diagnosis into spectral data analysis. Our results and novel insights suggest the following points of investigation for future work:
‚ In Chapter 6, we carried out an experiment in a controlled environment (screen house). We grew healthy plants and inoculated them with CBSD disease virus. Future experiments in this area would include more cassava varieties as well as studying on other diseases e.g CMD and CBSD.
‚ In a similar setup, future work would consider how crop nutritional deficiency caused by stress effects affects plant growth. This was a common question raised by our different stakeholders. To tackle this problem, an additional class to handle data from this group would be required during training.
8.1. Future work 89
‚ Our research should also be expanded by considering other crops, e.g maize, beans and rice. Initially, the same methods could be applied on dataset from the mentioned crops.
‚ Chapter 7 was our first attempt to the construction of a low-cost diagnostic device based on spectrometery data. We have identified several areas of potential improvements in our discussion.
‚ In some of our chapters we experimented with CNNs as a baseline. Being an area that has gained much popularity in recent studies, we envisage future studies in applicability of the neural networks.
This thesis serves as a basis for putting forward the efficient early detection of crop diseases using spectral information. Our research can contribute to an im-proved livelihood of a smallholder farmer in the Sub-Saharan Africa through in-creased crop yields and food security.
88 8. Summary and Outlook sava crop, once disease symptoms are visible in the aerial part of the plant, a lot of damage has been caused especially in the root of the plant which is majorly used as food. The novelty and usefulness of this research is evidenced in Chapter 6. The chapter presented results in using visible and near infrared spectral information to detect diseases in cassava crops before symptoms can be seen by the human eye. To test this hypothesis, we grew cassava plants in a screen house where they were inoculated with disease viruses and we monitored the plants over time collecting both spectral and plant tissue for wet chemistry analysis at each time step until the plants show disease. Our models in our case GMLVQ were able to detect cassava diseases one week after virus infection can be confirmed by wet lab chemistry, but several weeks before symptoms manifest on the plants.
Although the analysis of spectral data proved successful in our experiments, spectrometry devices are not commonly found on the market, ready to do auto-mated diagnosis of crop diseases. Also, most spectrometry devices require techni-cal knowledge to operate them, thus making them inapplicable for our end users (smallholder farmers). In our case, we purchased one expensive spectrometer (in 1000 USD) to achieve our research goals. In Chapter 7, we presented initial steps towards the development of a low-cost 3-D printed smartphone add-on spectrome-ter that can be used to diagnose crop diseases in the fields. The contribution of this chapter can be seen in the design of a diagnostic tool that is cheap and easy to use by smallholder farmers in developing countries in monitoring their crop status. The performance of the first prototype was not favourable, however, requirements for an improved version were stated.
8.1 Future work
The research presented in this thesis builds from previous studies as stated in the literature. The present work was transitioning from previous methods of diagnosis into spectral data analysis. Our results and novel insights suggest the following points of investigation for future work:
‚ In Chapter 6, we carried out an experiment in a controlled environment (screen house). We grew healthy plants and inoculated them with CBSD disease virus. Future experiments in this area would include more cassava varieties as well as studying on other diseases e.g CMD and CBSD.
‚ In a similar setup, future work would consider how crop nutritional deficiency caused by stress effects affects plant growth. This was a common question raised by our different stakeholders. To tackle this problem, an additional class to handle data from this group would be required during training.
8.1. Future work 89
‚ Our research should also be expanded by considering other crops, e.g maize, beans and rice. Initially, the same methods could be applied on dataset from the mentioned crops.
‚ Chapter 7 was our first attempt to the construction of a low-cost diagnostic device based on spectrometery data. We have identified several areas of potential improvements in our discussion.
‚ In some of our chapters we experimented with CNNs as a baseline. Being an area that has gained much popularity in recent studies, we envisage future studies in applicability of the neural networks.
This thesis serves as a basis for putting forward the efficient early detection of crop diseases using spectral information. Our research can contribute to an im-proved livelihood of a smallholder farmer in the Sub-Saharan Africa through in-creased crop yields and food security.
Bibliography
Abdullahi, I., Atiri, G. and Dixon, A.: 2003, Effects of cassava genotype, climate and the bemisia tabaci vector population on the development of African cassava mosaic geminivirus (acmv), Acta Agronomica Hungarica pp. 285–289.
Aduwo, J. R., Mwebaze, E. and Quinn, J. A.: 2010, Automated vision-based diagno-sis of cassava mosaic disease, Industrial Conference on Data Mining pp. 114–122. Alsin¸a, I., Duma, M., Dubova, L., Senberga, A. and Dagis, S.: 2016, Comparison of‘
different chlorophylls determination methods for leafy vegetables., Agronomy Research 14(2), 309–316.
Altman, N. S.: 1992, An introduction to kernel and nearest-neighbor nonparametric regression, The American Statistician, Vol. 46, [American Statistical Association, Taylor Francis, Ltd.], pp. 175–185.
Amanda, R., Kelsee, B., Peter, M., Babuali, A., James, L. and David, H. P.: 2017, Deep learning for image-based cassava disease detection, Frontiers in Plant Science
8, 1852.
Arens, N., Backhaus, A., D¨oll, S., Fischer, S., Seiffert, U. and Mock, H.-P.: 2016, Non-invasive presymptomatic detection of cercospora beticola infection and iden-tification of early metabolic responses in sugar beet, Frontiers in Plant Science
7, 1377.
Arias-Castro, E. and Donoho, D. L.: 2009, Does median filtering truly preserve edges better than linear filtering?, The Annals of Statistics 37(3), 1172–1206.
