Publication: Métodos de visión por computador para detección automática de líneas de cultivo curvas/rectas y malas hierbas en campos de maíz
Loading...
Files
Official URL
Full text at PDC
Publication Date
2018-10-24
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Complutense de Madrid
Citations
Exportar
Abstract
Cada día es mayor el uso de sistemas de visión por computador a bordo de vehículos autónomos para agricultura de precisión y su utilización en distintas tareas, demandando una atención especial. La discriminación entre cultivo y malas hierbas así como la identificación de las líneas de cultivo en imágenes obtenidas en campos de maíz (cultivo de surcos anchos) representan importantes retos, tanto desde el punto de vista de la aplicación de tratamientos selectivos como para un guiado preciso en la navegación de los mencionados vehículos. En cualquier caso, la calidad de las imágenes se ve afectada por las condiciones de iluminación no controladas en entornos agrícolas de exterior. Además, diferentes alturas y volúmenes de las plantas que se manifiestan por los distintos estados de crecimiento y la presencia de discontinuidades a lo largo de las líneas de cultivo debido a una mala germinación o defectos durante la siembra, dificultan los procesos de detección de líneas de cultivo y discriminación entre cultivo y malas hierbas. Las imágenes fueron tomadas bajo proyección de perspectiva con una cámara instalada a bordo del tractor y convenientemente colocada en la parte frontal. Con respecto a la detección de las líneas de cultivo, se han propuesto dos nuevos métodos para la detección de líneas curvas y rectas en campos de maíz durante los estados iniciales de crecimiento del cultivo y malas hierbas. El objetivo final es la identificación de las líneas de cultivo con dos propósitos: a) guiado preciso en vehículos autónomos; b) tratamientos específicos, incluyendo la eliminación de malas hierbas, situadas entre las líneas. Los métodos propuestos se diseñaron con la robustez requerida para abordar el problema de las condiciones adversas indicadas previamente y constan de tres fases consecutivas: (i) segmentación de la imagen, (ii) identificación de los puntos de comienzo de las líneas de cultivo y (iii) detección de las propias líneas. La principal contribución de estos métodos estriba en su capacidad para detectar líneas de cultivo curvas y rectas con espaciados regulares o irregulares entre las líneas, incluso cuando coexisten tipos de líneas en el mismo campo e imagen. Ambos métodos, difieren entre ellos en la fase de detección. Uno se basa en la acumulación de píxeles verdes y el otro en lo que se conoce como concepto de micro-ROIs (Region Of Interest). Los rendimientos de los métodos propuestos se compararon cuantitativamente frente a cinco estrategias existentes, consiguiendo precisiones entre el 86.3% y el 92.8%, dependiendo de si las líneas de cultivo son curvas o rectas con espaciado regular o irregular, con tiempos de procesamiento menores que 0.64 s por imagen...
The use of automatic machine vision systems onboard autonomous vehicles in precision agriculture is being widely used for different tasks demanding special attention. Crop/weed discrimination and crop rows detection in images captured from maize fields (wide row crops) represent important challenges for site-specific treatments and precise guidance for navigation purposes. This thesis addresses both problems with specific improvements with respect existing approaches. Image quality is affected by uncontrolled lighting conditions in outdoor agricultural environments. Also different plant (crops, weeds) heights, volumes and gaps along the crop rows due to lack of germination or defects during planting appear due to different growth stages affecting the crop/weed identification and the crop row detection process. The images were obtained under imaging perspective projection with a camera installed on board and conveniently arranged at the front part of a tractor. Regarding crop row detection, two new methods have been proposed for detecting curved and straight crop rows in images captured in maize fields during the initial growth stages of crop and weed plants. The final goal is the identification of the crop rows with two purposes: a) precise autonomous guidance; b) site-specific treatments, including weed removal, where weeds are identified as plants outside the crop rows. The proposed methods were designed with the required robustness to cope with the above adverse situations and consist of three linked phases: (i) image segmentation, (ii) identification of starting points for determining the beginning of the crop rows and (iii) crop rows detection. The main contribution of these methods is the ability to detect curved and straight crop rows having regular or irregular inter-row spacing, even when both row types coexist in the same field and image. Both approaches differ among them in the crop rows detection phase. One is based on the accumulation of green pixels and the other in the micro-ROI (Region Of Interest) concept. The performances of the proposed approaches were quantitatively compared against five existing strategies, achieving accuracies between 86.3% and 92.8%, depending on whether crop rows were straight/curved with regular or irregular spacing, with processing times less than 0.64 s per image...
The use of automatic machine vision systems onboard autonomous vehicles in precision agriculture is being widely used for different tasks demanding special attention. Crop/weed discrimination and crop rows detection in images captured from maize fields (wide row crops) represent important challenges for site-specific treatments and precise guidance for navigation purposes. This thesis addresses both problems with specific improvements with respect existing approaches. Image quality is affected by uncontrolled lighting conditions in outdoor agricultural environments. Also different plant (crops, weeds) heights, volumes and gaps along the crop rows due to lack of germination or defects during planting appear due to different growth stages affecting the crop/weed identification and the crop row detection process. The images were obtained under imaging perspective projection with a camera installed on board and conveniently arranged at the front part of a tractor. Regarding crop row detection, two new methods have been proposed for detecting curved and straight crop rows in images captured in maize fields during the initial growth stages of crop and weed plants. The final goal is the identification of the crop rows with two purposes: a) precise autonomous guidance; b) site-specific treatments, including weed removal, where weeds are identified as plants outside the crop rows. The proposed methods were designed with the required robustness to cope with the above adverse situations and consist of three linked phases: (i) image segmentation, (ii) identification of starting points for determining the beginning of the crop rows and (iii) crop rows detection. The main contribution of these methods is the ability to detect curved and straight crop rows having regular or irregular inter-row spacing, even when both row types coexist in the same field and image. Both approaches differ among them in the crop rows detection phase. One is based on the accumulation of green pixels and the other in the micro-ROI (Region Of Interest) concept. The performances of the proposed approaches were quantitatively compared against five existing strategies, achieving accuracies between 86.3% and 92.8%, depending on whether crop rows were straight/curved with regular or irregular spacing, with processing times less than 0.64 s per image...
Description
Tesis inédita de la Universidad Complutense de Madrid, Facultad de Informática, Departamento de Ingeniería del Software e Inteligencia Artificial, leída el 12/06/2017.