Low-altitude remote sensing and image processing for weed and disease monitoring by an unmanned radio-controlled helicopter | |
| Author | Grianggai Samseemoung |
| Call Number | AIT Diss. no.AE-11-01 |
| Subject(s) | Robots--Control systems--Remote sensing Weed control--Remote sensing |
| Note | A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Engineering in Agricultural Systems and Engineering |
| Publisher | Asian Institute of Technology |
| Abstract | Timely evaluation of crop growth, weed control and pest or diseases infections are extremely important for controlling the spread and thus preventing crop productivity losses. Near real-time images for quick assessment of the crop and weed status provide enough time for preventive measurement. A dedicated image data acquisition system and its supporting image processing algorithm is developed, which is evaluated in fields by using a tractor driven crane-attachment and an unmanned radio-controlled helicopter mounted low-altitude remote sensing (LARS) platforms. Evaluation is done in three parts. In the first part, the LARS images from a tractor driven crane-attachment were processed to compare the camera qualities at different altitude levels (5, 10 and 15m) for different crop maturity stages (7, 14, 21 and 28DAGs) in terms of crop growth and weed control in soybean field. In the second part of evaluation, the LARS images were also taken in similar conditions from an unmanned radio-controlled helicopter-attachment. The images obtained from the tractor driven crane-attachment and unmanned radio-controlled helicopter-attachment were also compared and validated with ground truth data in soybean field. In the third and final evaluation part, the LARS images from unmanned radio-controlled helicopter-attachment were taken in oil palm field and processed to distinguish infested trees from healthy trees; which were further analyzed by vegetation indices and validated with ground truth data. Performances of using true color digital (R-G-B) camera photography and color-infrared (CIR) photography (G-R-NIR) in the developed LARS system to acquire geo-referenced images were compared. The image processing analysis software (IPAS) created through this system found to be adequately representing crop and weed parameters for crop monitoring. Evaluation of the percentage of greenness in terms of crop growth revealed that NIRC (Color-infrared (CIR) photography based on tractor driven crane-attachment) was suitable for capturing images at all crop growth stages. Percentage of greenness from NIRC and RGBC (True color digital camera photography based on tractor driven crane-attachment) increased as altitude levels and crop maturity increased. Percentage of weeds in terms of weed monitoring from NIRC and RGBC increased as altitude levels and crop maturity increased. The quality of LARS images created by image processing software was acceptable and found suitable in terms of crop growth and weed density detections. Finally, LARS images from NIRC found more suitable as image qualities increase with altitude levels and DAG than RGBC. Moreover, NIRC made the system more flexible in terms of system integration, specific applications and cost attractiveness than RGBC. In order to process the LARS images from a tractor driven crane-attachment and unmanned radio-controlled helicopter mounted in soybean field, crop growth and weed infestation in soybean field were monitored by processing the LARS images taken from a crane-mounted and an unmanned radio controlled helicopter-mounted platforms. Images were taken for comparison between true color digital (R-G-B) and color-infrared (CIR) digital photography (G-R-NIR) cameras, fixed on a tractor-attached crane-mounted platform (RGBC and NIRC respectively), acquired at different heights. Similarly, true color digital camera and CIR digital camera were also fixed on an unmanned radio-controlled helicopter mounted platform (RGBH and NIRH respectively) to acquire images of the same field at different heights. All LARS images were processed to estimate vegetative-indices. These indices are used to distinguish the stages of crop growth and to estimate the weed density using ENVI software, and are validated using ground truth data. LARS images from the crane-mounted platform (image acquisition with less dynamic effects) and LARS images from the helicopter-mounted platform (image acquisition with considerable dynamic effects) are evaluated at different heights. It is found that RGBC and NIRC (crane-mounted platform) captured better quality images at lower altitude levels (<10m). This makes the crane-mounted platform as an attractive option in terms of system integration, specific low altitude applications and cost reduction. Whereas, RGBH and NIRH (helicopter-mounted platform) are found suitable at altitude levels >10m. Comparison of NDVIcrane-mounted (NDVI based on the crane-mounted LARS images) and NDVIhelicopter-mounted (NDVI based on the helicopter-mounted LARS images) is made for different altitude levels and crop growth stages. It is found that NDVI values taken at 28 days after germination (DAG) showed a strong relationship with the altitude levels, attaining a coefficient of determination (R2) of 0.75 for NDVIcrane-mounted and 0.79 for NDVIhelicopter-mounted. However, high altitude levels (>10m) decreased NDVI values for both the crane and the helicopter-mounted systems. Higher R2 values (≥0.7) were also obtained for indices estimated from crane-mounted images (NDVIcrane-mounted) and from helicopter-mounted images (NDVIhelicopter-mounted) with the index obtained using a ground spectrometer (NDVISpectro), which shows an adequate suitability of the proposed LARS platform systems for crop growth and weed infestation detection. Furthermore, chlorophyll content was also well related with the indices from crane and helicopter-mounted images with high R2(>0.75) for 7, 14, 21 and 28 DAGs. To monitor pest infestation in the oil palm field, the radio-controlled helicopter-mounted LARS platform was used. The acquired LARS images were processed to estimate vegetative-indices and thereby detecting upper stem rot (Phellinus Noxius) disease in both young and mature oil palm plants. The indices helped discriminate healthy and infested plants by visualization, analysis and presentation of digital imagery software, which were validated with ground truth data. Good correlations and clear data clusters were obtained in characteristic plots of NDVIhelicopter-mounted and GNDVIhelicopter-mounted against NDVISpectro and chlorophyll content, by which infested plants were discriminated from healthy plants in both young and mature crops. The chlorophyll content values (μmol m-2) showed notable differences among clusters for healthy young (972-1100), for infested young (253-400), for healthy mature (1210-1500), and for infested mature (440-550) oil palm. The correlation coefficients (R2) were in a reasonably acceptable range (0.62-0.88). The vegetation indices based on LARS images, provided satisfactory results when compared to other approaches. The developed technology showed promising scope for medium and large plantations. The adoption of the image data acquisition systems with low altitude remote sensing (LARS) platforms and precision agriculture for small and medium farm holdings in developing countries was verified and recommended for quick implementation for better profits. |
| Year | 2011 |
| Type | Dissertation |
| School | School of Environment, Resources, and Development (SERD) |
| Department | Department of Food, Agriculture and Natural Resources (Former title: Department of Food Agriculture, and BioResources (DFAB)) |
| Academic Program/FoS | Agricultural and Food Engineering (AE) |
| Chairperson(s) | Soni, Peeyush;Jayasuriya, H.P.W. |
| Examination Committee(s) | Manukid Parnichkun;Shivakoti, Ganesh P.;Salokhe, Vilas M.;Roongruang Kalsirisilp;Noguchi, Noboru; |
| Scholarship Donor(s) | Rajamangala University of Technology Thanyaburi RMUTT, Thailand; |
| Degree | Thesis (Ph.D.) - Asian Institute of Technology, 2011 |