Rice leaf blast spot segmentation method based on hyperspectral imaging technology and its spectral characteristics analysis

Rice blast is one of the most serious rice diseases in the northern and southern rice-growing areas of my country. It is known as one of the three major rice diseases

along with sheath blight and bacterial leaf blight. At present, the identification of rice blast is mainly done manually through picture comparison or text description.

However, these identification methods are too subjective, require high professional quality of workers, and are inefficient. They often cause errors in human judgme

nt, which makes it difficult to accurately and timely prescribe the right medicine, thus affecting the prevention and control effect and causing rice production reduc

tion. Hyperspectral imaging technology is a new technology that organically combines traditional imaging technology and spectral technology. Imaging technolo

gy can be used to obtain image information of crops, and spectral technology can be used to obtain spectral information of crops.

This paper takes rice leaves with rice blast spots as the research object, uses a hyperspectral imaging system to obtain its hyperspectral image, and then uses the

principal component analysis method to determine the principal component image suitable for spot segmentation. Finally, the density segmentation method is

used to realize the identification of rice leaf blast spots. On this basis, the spectral change law of the spot part and its spectral difference with normal leaves are

analyzed.

1. Materials and methods

1.1 Experimental materials

The rice variety used in the experimental study is Guanglu Ai No. 1, which is a susceptible variety. After soaking and germination, a potted test was conducted

with 60 pots under the same management level. When the rice seedlings grew to the 3rd to 4th leaves, the rice blast pathogen was sprayed in an artificial inoc

ulation box (shake the prepared spore suspension and spray it evenly on the leaf surface until the leaves are completely covered with small water droplets). After

inoculation, the rice seedlings were moved into a dark box for 24 hours of moisturizing and then transferred to a greenhouse for moisturizing culture to promote

disease. Hyperspectral image data was collected for 5 consecutive days to obtain samples with different disease levels.

1.2 Hyperspectral imaging system

This study used a 400-1000nm hyperspectral camera, and the FS13 product of Hangzhou Caipu Technology Co., Ltd. can be used for related research. The spectral

range is 400-1000nm, the wavelength resolution is better than 2.5nm, and up to 1200 spectral channels. The acquisition speed can reach 128FPS in the full spectrum,

and up to 3300Hz after band selection (supports multi-region band selection).

1.3 Principal component analysis of rice leaf blast hyperspectral images

Hyperspectral imaging technology can obtain many very narrow spectral continuous image data, which have strong inter-spectral correlation. Principal component

analysis is mainly a method for transforming a set of interrelated data into a set of unrelated variables through orthogonal transformation. It can fully remove the

correlation and concentrate the useful information into the fewest principal components. The principal component (PC) band is a linear synthesis of the original

bands, and they are unrelated to each other. The first principal component contains the largest data variance, the second principal component contains the second

largest variance, and so on. The last principal component band is displayed as noise because it contains a very small variance. Through principal component analysis,

the principal component image can be selected according to the size of the contribution rate. Table 1 shows that the contribution rate of the first 6 principal

component images reached 99.81%. Figure 2 shows the PC1 to PC6 principal component images of rice leaf blast. Comparing these 6 principal component images, the

PC2 image lesions are significantly different from the leaf background, which is conducive to the segmentation of rice leaf blast lesions.

2. Results and Analysis

2.1 Image Segmentation Results

Figure 3 is a binary image after PC2 image segmentation. The white part represents rice leaf blight spots. It can be seen that the binary image basically reflects the

entire lesion area of the rice leaf under the stress of leaf blight. In order to further test the effectiveness of the segmentation method, principal component analysis

and density segmentation method were applied to 60 samples in this experiment. Among them, the lesion part in 54 images can be accurately segmented, and the

segmentation rate is 90%.

2.2 Analysis of spectral characteristics of lesions

The segmented binary image is multiplied with the original hyperspectral image to obtain a hyperspectral image containing only the lesion area. This area is selected

as the region of interest and its spectral characteristics are analyzed. Figure 4 shows the spectral curves of the normal part and the region of interest of the leaf blight

lesion. It can be seen from the lesion curve that the lesion spectral curve generally shows an upward trend in the 400-988nm band. In the blue light band (435-480nm),

the reflectivity is low, but it is rising until the green light band. In the red light band (600-760nm), the reflectivity gradually increases, and then begins to decline at

around 650nm, but the decline is small, and then rises, forming a shallow trough at around 680nm, and rising steeply from 690 to 760nm. In the near-infrared band,

the spectral reflectivity rises slowly after 760nm until it reaches the highest reflectivity value at 988nm. By comparing the spectral reflectance of rice leaf blast lesions

and normal areas in the visible-near infrared range, it was found that the reflectance of rice leaf blast lesions was greater than that of normal leaf areas in the blue

light band (435-480nm) and red light band (600-700nm). A shallow trough appeared in the rice leaf blast lesion curve near 680 nm in the red light band, where the

spectral reflectance of the leaf blast lesions was significantly different from that of the normal leaf areas. In the green light band (530-580nm), the spectral reflectance

of the rice leaf blast lesions was slightly lower than that of the normal leaf parts, which was mainly related to the gradual decline of chlorophyll and the gradual

appearance of xanthophyll in rice leaves during this period. In addition, in the near-infrared band (720-988nm), the spectral reflectance of the diseased part of the rice

leaf decreased compared with that of the normal leaf part. This change was determined by the internal cell structure of the leaf.

3. Conclusion

This paper uses hyperspectral imaging technology to detect rice leaf blast. The conclusions are as follows:

(1) The principal component analysis method is used to reduce the dimension of hyperspectral data, and the PC1 to PC6 principal component images are obtained.

The second principal component image is selected to segment and identify rice leaf blast, and the lesion segmentation rate is 90%.

(2) The spectral characteristics of the rice leaf blast lesions and the normal leaf areas of interest are analyzed. In the green light band (530~580nm) and the red light

band (600~700nm), the spectral reflectance of the rice leaf blast area shows a decreasing and increasing trend compared with the normal leaves. In the near-infrared

spectrum, rice leaves in the near-infrared band (720~988nm) will cause the spectral reflectance of the rice leaves to decrease after infection.

(3) Through the research of this paper, the accurate segmentation of rice leaf blast lesions on a single leaf is achieved, which provides a basis for the next step of

field application of multispectral/hyperspectral imaging technology to detect the occurrence of rice leaf blast in the rice canopy.