This presentation description is as outlines in the abstract below

Elijah Cheruiyot¹, Arnold Bett², Adolfo Posadas¹, Dieudonné Harahagazwe¹, Hildo Loayza³, Susan Palacios³, Mario Balcázar³, Luis Silva³ and Roberto Quiroz³

¹International Potato Center, Nairobi, ²University of Nairobi, ³International Potato Center, Lima

Crop statistics in agriculture are important tools for planning, policy making and timely interventions to address food insecurity. Conventional methods for gathering crop statistics are costly, not frequently updated and often inaccurate. The potential of satellite remote sensing in gathering crop statistics data has been demonstrated, but associated costs are also prohibitively high and the data quality is often negatively affected by clouds. The objective of this project is to use Unmanned Aerial Vehicle (UAV)-based remote sensing technologies to provide a cost effective means of gathering, processing and interpreting adequately accurate and timely crop statistics data at a large scale with minimal effect of clouds. In order to meet this objective, low-cost UAV platforms must be coupled with high quality sensors and sophisticated-yet-user-friendly data processing techniques. The approach under investigation is as follows: 1) Locally assemble and test UAVs; 2) Construct a spectral library of crops and other land cover types in the target region; 3) Acquire aerial images and process them to identify crops and estimate cropping areas; 4) Upscale the results to cover wider regions by fusion with satellite data. As of today, one UAV has been assembled in Nairobi and tested in Tanzania, through a process that not only reduced cost but also built local capacity in assembling and maintenance of the UAV. Two data processing programmes have been developed at CIP, namely ISAM_CIP for image stitching and Spectra-CIP for collection of spectral measurements with spectroradiometer, while design of improved sensors for various agronomic applications is on-going. Preliminary results obtained from data collected in the district of Misungwi, Tanzania show that different crops can be easily identified in an image by means of classification based on optical characteristics collected separately using spectroradiometer on one hand, and on the other based on texture. Texture-based classification seems to be particularly suitable for less experienced users as it does not require additional knowledge and equipment to collect the input spectral characteristics data necessary for spectral-based classification. Furthermore significant progress has been made in scaling up these results through fusion of fine resolution but spatially limited UAV data with coarser resolution satellite data. However further work is required to improve the accuracy of texture-based classification methods in order to allow crop identification from images taken with regular cameras.