Mapeamento da cobertura de plantas daninhas utilizando imagens digitais e geoestatística

• Autor(a) principal: Silva Júnior, Mário Cupertino da
• Data de Publicação: 2010
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: LOCUS Repositório Institucional da UFV
• Texto Completo: http://locus.ufv.br/handle/123456789/690
• topic: Plantas; Solo; Plants; Soil; CNPQ::CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA

The objective of this study was to map weed percent coverage using techniques of digital image processing and geostatistics in row crops. Two experiments were performed. The first was conducted in a 0.8 hectare experimental area belonging to the Universidade Federal de Viçosa (UFV) in the city of Coimbra, MG, Brazil, whose area was irrigated with a central pivot system. The area was planted with common beans, Ouro Vermelho cultivar, where half of the area was crop in tillage system and the other half in no-tillage. The second experiment was conducted on a private property of approximately 1.2 hectares cultivated with sunflower, with no irrigation system, located in Aguilar de Bureba, Burgos province, Spain. In the first study, a machine vision system was built, with two digital cameras and a DGPS(Trimble Pathfinder Pro XRS) set up on the central pivot structure. The central pivot moved to sample the entire area of study. One camera acquired color images in the visible bands (RGB), and the other acquired images in the near infrared band (NIR). These cameras captured images simultaneously of the same scene. The scene coordinates were acquired by the DGPS, and it was assumed to be in the center of the common area of the two images. These images were acquired in a grid pattern.The images were processed for the percentage of weed cover estimation. Once it was acquired all the georeferenced weed percentage values, it was possible to construct maps using geostatistical techniques. The system performance was access for the two cameras and for the two tillage systems. The maps generated by using color images were more reliable than those using NIR images for weed infestation detection in both tillage systems since they presented a better contrast between plants and background. In the second experiment, the system used two different digital color cameras, with different image resolutions, and a GPS. All equipments were set on a tractor, simulating height of a central pivot. The two cameras captured color imagesof the same scene and at the same height, but with different spatial resolutions. It was again assumed that the GPS coordinates referred to the image center. Sample images of the area were acquired in a grid pattern at camera heights of 3 and 4 m, and at 37 and 46 days after planting (DAP). The images were processed for weeds infestation estimation for each position. Thus, maps for camera, height and date were built by the developed system. The higher resolution camera was presented better performance to map the percentage of weed coverage and identify variability of these plants in the area under study, at both heights and growth stages since it presented better contrast between plants and soil than the other camera. The maps at both stages of growth presented similarity for both tested heights and cameras.