Research On Detection Methods Of Plant Disease Using Computer Vision

Early detection of plant disease based on computer vision technique can help us to prevent thespread of plant disease effectively and stop the disease in the beginning. It can not only improve theyield and quality of crops, but also reduce the dosage of pesticide. Therefore the application prospect ofthis research is very attractive. The main research contents and results were as follows:1. The research advancements and achievements in the field of detecting plant disease and plantphysiological status using computer vision technique were reviewed. The existing problems andrequirements were put forward.2. Three research systems were set up. The visible light image grabbing system was composed ofsix fluorescent lamps, a CCD camera(TMC-7DSP(PULNIX)CCD), a frame grabber (MeterⅡ/MC(Matrox.Inc))and so on. The infrared thermal imaging system was composed of a thermal imager(Talisman K90C (ISG.Inc)), Sony recorder, a frame grabber and so on. The multi-spectral imagingsystem was composed of a 3-CCD camera (MS3100 Duncan 3CCD), a frame grabber (PCI-1424) andso on.3. Visible light images of plant disease were acquired and pre-processed. The images weresegmented with an indicator composed of R, G, B values. The wrong pixel information which occurredduring background segmentation was recovered in this research. Analyzing the color characteristics ofdisease spots, G/R and G/B were used to segment the disease spots on leaves and got a satisfied result.In this research, plant infection severity ratio was determined by image processing technique. Compared tothe traditional method, the result showed that computer vision technique is a feasible way to determine plantinfection severity ratio.4. In order to analyze the infection situation of plant and thermal images sufficiently, experimentslike electron microscopy trail, chlorophyll contents detection trail etc. were also done in this study. Theinfected leaves showed a pre-symptomatic decrease in leaf temperature about 0.5-1.3℃lower than thatof the healthy leaves. The temperature difference allowed the discrimination between the infected andhealthy leaves before the appearance of visible necrosis on leaves by thermal imaging.5. Multi-spectral imaging system was also set up for this research. The multi-spectral images werecompared with the visible light images and thermal images. The result showed that the wave bands ofthe 3-CCD camera used in this research were not sensitive to early plant disease. So the multi-spectralimages couldn't detect the plant disease before the appearance of visible necrosis on leaves.6. In this research, a new wavelet-based approach was proposed for thermal images segmentation.The results showed that multi-resolution of wavelets is able to detect objects at multi-scale. It canremove the background more exactly and easily than conventional segmentation methods.