Hybrid Segmentation – Artificial Neural Network Classification of High Resolution Hyperspectral Imagery for Site-Specific Herbicide Management in Agriculture

Site-Specific Herbicide Management (SSHM) in Precision Agriculture (PA) requires weed detection in crop fields for directed herbicide application instead of spraying entire fields. This has significant economic and environmental advantages for improved agricultural practices that are essential given forecast increases in global population and food production needs. In this study, a new hybrid segmentation Artificial Neural Network (HS-ANN) method was compared to standard Maximum Likelihood Classification (MLC) for improving crop/weed species discrimination in SSHM/PA. Very high spatial resolution (1.25 mm) ground-based hyperspectral image data were acquired over field plots of canola, pea, and wheat crops seeded with two weed species (redroot pigweed, wild oat) in southern Alberta, Canada. The very high spatial and spectral resolution image data required development of a simple yet efficient vegetation index (Modified Chlorophyll Absorption in Reflectance Index (MCARI)) threshold segmentation to separate vegetation from soil for classification. The HSANN consistently outperformed MLC in both single date and multi-temporal classifications. Higher class accuracies were obtained with multi-temporally trained ANNs (84 to 92 percent overall), with improvements up to 31 percent over MLC. With advancements in imaging technology and computer processing speed, this HS-ANN method may constitute a viable farm option for real-time detection and mapping of weed species for SSHM in agriculture. Introduction Site-Specific Herbicide Management (SSHM) involves selectively applying herbicides to an agricultural field based on identified zones of weed density rather than spraying an entire field (Thompson et al., 1991). As a key Hybrid Segmentation – Artificial Neural Network Classification of High Resolution Hyperspectral Imagery for Site-Specific Herbicide Management in Agriculture P.R. Eddy, A.M. Smith, B.D. Hill, D.R. Peddle, C.A. Coburn, and R.E. Blackshaw component in Precision Agriculture (PA), SSHM can provide substantial benefits through reducing the amount of herbicide required for weed control and crop protection since the weed-controlling chemical is only applied where it is actually needed (Brown and Steckler, 1995; Medlin et al., 2000; Blackshaw et al., 2006). Techniques for implementing SSHM strategies are of increasing importance for compliance with strict environmental regulations and are also advantageous economically. SSHM techniques may result in a 30 to 72 percent reduction in herbicide requirements (Mortensen et al., 1995) and considering that global herbicide product sales totalled $14.8 billion (USD) in 2006 (Crop Life, 2007), could constitute a substantial savings to producers. This reduction of chemicals applied also reduces the risk of environmental contamination as a result of ground-water leaching and introducing less chemicals into the atmosphere (Lindquist et al., 1998; Radhakrishnan et al., 2002; Smith and Blackshaw, 2003). With projections of global population increases in the coming years and the associated increased reliance on agriculture to meet challenging food production demands (Tweeten, 1998; FAO, 2007), effective and efficient agricultural practices such as SSHM/PA will be crucial to reducing environmental impacts and helping ensure the economic viability of agricultural systems. Operational implementation of real-time SSHM requires on-board image acquisition and processing systems and precise control of herbicide spray applicators (Tang et al., 1999; Brown and Noble, 2005). The image acquisition and processing system must rapidly differentiate weeds from crop (Hutto et al., 2006; Grey et al., 2007) and provide the sprayer control with a map of weed location and density in near real-time. This map is built up in the field with immediate herbicide application dependant on Artificial Intelligence (AI) system decision making. Such a system requires accurate species recognition as well as computational efficiency (Tian et al., 1999; Tang et al., 2000; Burks et al., 2000b). The rich information provided by hyperspectral sensor systems requires efficient data processing and interpretation tools, such as AI methods, for practical application to real-time SSHM. Artificial Neural Networks (ANNs) are uniquely suited to these image processing tasks and can handle complex feature space and integrate different data types (Atkinson and Tatnall, 1997). As a nonparametric approach they offer significant PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Oc t obe r 2008 1249 P.R. Eddy is with the Alberta Terrestrial Imaging Center, 401, 817-4 Ave. South, Lethbridge, AB, T1J 0P3, Canada; and formerly with the Department of Geography, University of Lethbridge, and Agriculture and Agri-Food Canada, Lethbridge, AB, Canada ([email protected]). A.M. Smith is with Agriculture and Agri-Food Canada, 5403-1 Ave. South, Lethbridge, AB, T1J 4B1 Canada, and the Department of Geography, University of Lethbridge, Lethbridge, AB, T1K 3M4 Canada. B.D. Hill and R.E. Blackshaw are with Agriculture and Agri-Food Canada, 5403-1 Ave. South, Lethbridge, AB, T1J 4B1 Canada. D.R. Peddle and C.A. Coburn are with the Department of Geography, University of Lethbridge, 4401 University Drive West, Lethbridge, AB, T1K 3M4 Canada. Photogrammetric Engineering & Remote Sensing Vol. 74, No. 10, October 2008, pp. 1249–1257. 0099-1112/08/7410-1249/$3.00/0 © 2008 American Society for Photogrammetry and Remote Sensing SA-AI-05.qxd 11/9/08 9:29 AM Page 1249