This thesis presents the development and testing of a compact multispectral camera system designed for use in precision agriculture and agricultural robotics. The system operates on the principle of sequential exchange of bandpass spectral filters mounted on a rotating filter wheel positioned in front of the optical sensor, enabling the acquisition of images across thirteen distinct spectral bands spanning the visible, ultraviolet, and near-infrared regions of the electromagnetic spectrum.
The hardware comprises a filter wheel, a stepper motor for driving the filter wheel, an absolute position encoder for the stepper motor, a monochromatic image sensor paired with a lens for image acquisition, an STM32L412KB microcontroller for handling low-level tasks, a Raspberry Pi 4 embedded computer for managing high-level tasks, and a custom-designed printed circuit board for straightforward and organised interconnection of the aforementioned components.
A software library was developed in Python 3, operating on a client-server model and enabling wireless communication over a local Wi-Fi network. The library provides a simple programming interface for the acquisition and transfer of spectral images to a remote computer, calibrated and ready for further processing.
The functionality of the device was validated by computing the NDVI and SAVI vegetation indices across multiple scenes containing different types of vegetation in a real outdoor environment. The results demonstrated clear differentiation between healthy vegetation, dying plants, and non-vegetative objects, with the SAVI index proving more robust in environments with greater soil exposure.
In the course of the research, certain limitations of the system were also identified, in particular the need for manual adjustment of focus and exposure time, a suboptimal wide-angle lens, and the possibility of Wi-Fi network dropouts during device operation. Proposals for potential solutions to the identified limitations are also provided.
|
