The goal of this research was to advance precision agriculture research by promoting machine vision technologies for real-time plant variability sensing and in-field application. Specifically, research efforts were focused on the development and implementation of engineering solutions for crop and weed sensing, variable-rate control, and vehicle positioning, the three fundamental components required for a machine vision system to function by the principles of precision agriculture.; A supervised color image segmentation scheme using a binary-coded genetic algorithm (GA) was developed for vegetation detection in hue-saturation-intensity (HSI) color space under outdoor lighting conditions. The results showed that this innovative segmentation scheme performed equally well as compared to other cluster analysis-based segmentation schemes.; A novel texture-based broadleaf and grass classification method was developed using a low-level Gabor wavelet filter bank-based feature extraction algorithm and a high-level neural network-based pattern recognition algorithm. A 100% classification accuracy was achieved when classifying the test samples from five weed species under real-time constraints.; To promote the adoption of the machine vision-based selective spraying technology, the sensing system cost was significantly reduced through the incorporation of the low-cost video-conferencing cameras. A prototype system was built and validated through experimentation.; To improve the accuracy of sprayer position and orientation estimation, a Kalman filtering sensor fusion technique was implemented to integrate the GPS system and wheel encoders. The developed positioning system reduced the position error by 80% as shown by evaluative tests, where machine vision was innovatively introduced to generate sub-centimeter accuracy validation tracks.; Through the addition of new hardware and the enhancement of the software functionality, the developed machine vision-based selective spraying system was further converted into a multifunctional platform for both real-time in-field variability mapping and selectively spraying.; In-field variations associated with corn plant spacing, growth stage, and population can lead to a significant yield differences. Since the ability to reduce these variations is directly related to the planter performance, a machine vision-based emerged corn plant sensing system (ECS) was developed for the performance evaluation for prototype planters. With the real-time image sequencing capability, the system also achieved an average spacing measurement error of less than 10 mm.
展开▼
