Research on Video Image Recognition Technology of Maize Disease Based on the Fusion of Genetic Algorithm and Simulink Platform

: In order to improve the segmentation accuracy of maize disease leaves with genetic algorithms and reduce segmentation time, this paper proposed a video image recognition technology of maize disease based on the fusion of genetic algorithm and Simulink simulation platform. The technology firstly uses Simulink simulation platform to process the real-time video data captured, including sharpening, segmenting and smoothing, to improve image clarity and quality; Secondly, it uses genetic algorithm to generate optimization model to determine the optimal image of maize diseases; Finally, it fuses genetic algorithms and Simulink platform to analyze and recognize these optimal images. The study results of maize big-spot disease images show that image grey scale values changes after the process of the fused optimal algorithm so that the characteristics of maize diseases are high lightened and the recognition rate of maize disease video image is improved remarkable. The algorithm provides a valid basis for the identification and the diagnosis and treatment of maize disease.


Introduction
Computer image processing technology is an important component in the field of artificial intelligence, and between man and computer basic theory and application technology provides a specific interface.But the application of image processing technology in agricultural engineering research starts late in our country, mainly in the crop disease diagnosis [1][2], agricultural product quality detection [3][4], crop growth status monitoring [5][6], agricultural crops intelligent classification [7], etc.The present study show that using image processing technology not only can detect, stem diameter, leaf area, leaf circumference petiole Angle of crops such as external growth parameters, can also with the fruit surface color and fruit size to judging the fruit maturity, and crop water lack of fertilizer, and so on and so forth [8].Computer image processing technology in crop production and research of information collection and has a large amount of information, high speed and high precision of significant characteristics and advantages, and solve some manual measurement is difficult to solve the problem.Domestic started relatively late in the image recognition and processing, but the study of foreign theory made certain optimization, also has obtained certain research results.More mature by identifying plants in the static image texture and color features combined with neural network to achieve for the identification of crop nutrient deficiency.Domestic only video image processing technology was applied to road traffic and dynamic video processing of events, but applied to crop disease monitoring information has not been reported to see.

The research method 2.1 Simulink
Simulink is a visualization simulation tool of MATLAB, it is a kind of block diagram design based on MATLAB environment, is to realize the dynamic system modeling, simulation and analysis of a software package that is widely used in linear systems, nonlinear systems, digital control and digital signal processing (DSP) in the modeling and simulation.Simulink continuous sampling time and discrete sampling time can be used or two mixed sampling time, it also supports multi-rate system, also is the system of the different parts have different sampling rate.In order to create the dynamic system model and Simulink provides a model block diagram of the graphical user interface (GUI), the creation process can complete just click and drag the mouse operation, it provides a more rapid and straightforward way, and users can immediately see the results of simulation of the system.

Genetic algorithms
Genetic algorithm is also in the field of computer science and artificial intelligence to solve the optimization of a heuristic search algorithm, is a kind of evolutionary algorithms.This heuristic is often used to generate useful solutions to optimization and search problems.Evolutionary algorithm was originally borrowed some phenomenon in evolutionary biology and developed, these phenomena, including heredity, mutation, natural selection and hybridization, etc. Genetic algorithm (ga) in the case of wrong selecting fitness function is likely to converge to local optimum [1], and cannot achieve the global optimal.
In genetic algorithm, the above several characteristics together in a special way: based on the parallel search of chromosome group, with a nature of speculation selection operation, switching operation and mutation operation.This particular combination differentiate between genetic algorithm with other searching algorithm.

Video analyses
Video analysis, IVS (Intelligent Video System), Video analysis technology is the use of computer image analysis technology vision, through the background and target separation and analysis in the scene and track targets within the camera scene.Video content analysis technology based on visual surveillance camera video image analysis, and has the capability of the background to the variety of the filter, by establishing the model of human activity, with the help of the computer's high speed computing power using a variety of filters, ruled out monitoring scene African human interference factors, accurate judgment in video surveillance images in various human activities.

Image classification
Color is a visual features on the surface of the object, the color of each object has its own particular characteristics, to take some similar color to the same object characteristics, so we can according to the characteristics to distinguish object.The color with color in a feature in image classification can be traced back to the Swain and Ballard color histogram method is put forward.Due to the size of the color histogram is simple and with image, rotation changes not sensitive, got the attention of researchers is widely, now almost all image database system based on the content classification to classify color classification method as an important means, and put forward many improved methods, summarized the main can be divided into two categories: the global color feature index and local color feature index.

The genetic algorithm with Simulink platform integration of video image recognition technology of corn 3.1 Data mining based on genetic algorithm
Using the matlab software platform to maize disease image for processing object, by using the genetic algorithm coding, selection, crossover and mutation in the four basic programming operation, find out the objective function, according to the objective function to find the fitness, according to the fitness retains the image of the optimal selection as the candidate solution, give up the other candidate solutions.After iterative cycle, choose the biggest fitness individual images as the optimal solution output.The specific process as shown in figure 1.

SUMILINK platform for data processing
Simulink is a visualization simulation tool of MATLAB, it is a kind of block diagram design based on MATLAB environment; realize the dynamic system modeling, simulation and analysis of a software package.In the MATLAB/SUMILINK Video and Image Processing Block set module library can also be used for the Image Processing.
First of all, establish related model, so that the image is processed with the facts, as shown in figure 2.  Before more complex image processing, need to convert the image first, according to the requirements of each image is converted to the required type; Images of geometric transformation is the image in the size, location and shape of transformation, the collected in the image rotation, translation, mirror to determine the optimal image, in order to deal with.Figure 3 for corn big spot without image processing:  , edge detection operator is used the image of vertical and horizontal difference gradient approximation operator, namely: In edge detection, the calculation f  of each pixel in the image, then the absolute value, the final threshold operation can be achieved.Robert operator calculation formula is: Roberts operator is composed of the following two templates: The Roberts operator for processing image is shown in figure 6 Is a kind of first order differential operator edge detection, using pixel point up and down, left and right neighboring points of gray level difference, at the edge The extreme edge detection, remove part of the false edge, has a smooth function to noise.Its principle is in the image space by using two direction template and image neighborhood convolution to complete, the two direction template a horizontal edges detection, a vertical edge detection.For digital images ) , ( y x f , Prewitt operator are defined as follows: Prewitt operator is composed of the following two templates: The Prewitt operator for processing the image shown in figure 7: Each pixel of the image of the transverse and longitudinal gradient approximation can be used in combination with the following formula, to calculate the size of the gradient.Hysteresis threshold need two threshold -high threshold and low threshold.Assuming that the image is the important edge in the continuous curve, so that we can track the given the fuzzy part of the curve, and there will be no composition curve to avoid the noise pixels as edges.So we start with a large threshold, which will identify real edge we're pretty sure, using the export in the direction of the information, we started from the real edge at the edge of the image tracking the whole.At the time of tracking, we use a smaller threshold, so that you can trace curve of the fuzzy part until we go back to the beginning.
Once this process is complete, we will get a binary image, each point of said is an edge point.
A achieve sub-pixel accuracy improvement implementation is in the direction of the gradient on the edge of the detection of the second order directional derivative to zero.Using the above methods of corn after the big spot gray level image processing, test results as shown in figure 12, shown in figure 13 Figure 12 Gray level and direct gray-scale transformation after the original image From the multiple line chart 3.1, using the gray scale transformation technology of direct gray-scale transformation method after processing the image grey value is decreased obviously, and reduce useless blurs the image grey value, instead, to enhance the image resolution highlighted the disease characteristics, is conducive to the identification of diseases.
Figure13 Gray original image and the image after the corrosion From the point of double line chart 13, using corrosion technology processing after image grey value increase.With direct gray-scale transformation processing figure, to the contrary, but corrosion technology can accurately extract the characteristics of large spot disease of maize, to facilitate the identification of diseases.

Results and Discussion
(1) Using the Simulink platform, the algorithm established the reorganization model of maize disease video images, improved image resolution, high lightened disease characteristics and increased the recognition rate of maize diseases.
(2) This paper fused genetic algorithms and Simulink platform effectively and built a model with remote video image analysis and classification reorganization of maize diseases.It gained the texture, color and geometry characteristics of maize diseases through field video image analysis and then established analytical model for all kinds of maize diseases.It monitored effectively the real-time video images of field diseases, processed the video image captured by sharpening, segmenting and smoothing and improved image clarity and quality; It fused genetic algorithms and Simulink platform to resolve and identify the optimal image.Results have shown that image grey scale values changes after the process of the fused optimal algorithm, high lightens the characteristics of maize diseases are high lightened, improves the recognition rate of maize disease video image remarkable and provides a valid basis for the identification and the diagnosis and treatment of maize disease.

Figure 1
Figure 1 Genetic algorithm process

Figure 2
Figure2The image processing model(1) the image transformation and geometric transformation Before more complex image processing, need to convert the image first, according to the requirements of each image is converted to the required type; Images of geometric transformation is the image in the size, location and shape of transformation, the collected in the image rotation, translation, mirror to determine the optimal image, in order to deal with.Figure3for corn big spot without image processing:

Figure 5
Figure 5 Rotate30° the figure corn big spot (2) The signal processing module library Analysis and enhanced module library (Analysis Enhancement) of Block Matching ), it can be collected by the image sequence or video frames used for motion estimation, also used to remove redundant information, between the video frames in video compression; Edge Detection, Edge Detection), you can use the Sober operator, Roberts operator, Prewitt operator or Canny operator to find objects in the image Edge, reduce the amount of data, and eliminated can think irrelevant information, retained the image important structural properties; Conversion module library (Conversions) automatic threshold (Auto threshold) can be obtained in the gray image into binary image.The use of the rest of the sons of several module library module of image processing.a)Roberts operator edge detection For discrete image ) , ( y x f

Figure 7
Figure 7 Prewitt operator edge detection figure


And then can use the following formula to calculate gradient direction.By the Sobel operator for processing the image shown in figure8:

Figure 8
Figure 8 Sobel operator edge detection figure d) The Canny operator edge detection Canny use of hysteresis threshold.Hysteresis threshold need two threshold -high threshold and low threshold.Assuming that the image is the important edge in the continuous curve, so that we can track the given the fuzzy part of the curve, and there will be no composition curve to avoid the noise pixels as edges.So we start with a large threshold, which will identify real edge we're pretty sure, using the export in the direction of the information, we started from the real edge at the edge of the image tracking the whole.At the time of tracking, we use a smaller threshold, so that you can trace curve of the fuzzy part until we go back to the beginning.Once this process is complete, we will get a binary image, each point of said is an edge point.A achieve sub-pixel accuracy improvement implementation is in the direction of the gradient on the edge of the detection of the second order directional derivative to zero.

Figure 11 Corn big spot complement graphs 3 . 4
Figure9 Canny operator edge detection figure(3) image enhancement of Simulink implementation Hd camera and remote monitoring system for video quality may not be used, so in video image enhancement is conducive to further analysis of the image.Image smoothing to highlight the image of wide area, the low frequency part, trunk or suppress image noise and interference of high frequency component processing, make the image brightness flat gradient, gradient decrease mutations, improve image quality.Image sharpening is contrary to the image smooth, used for compensating the contour image, enhancing image edge and gray level jump part, makes the image becomes clear.To combine two methods of processing, can improve the quality of the image better for analysis.a)The image binarization processing is to the point of the image gray level set to 0 or 255, or the entire image showing a clear black and white effect.About 256 brightness levels of gray image through appropriate threshold selection and can still reflect the overall image and local features of binary images.The binary processing of image is shown in figure10: