Weeds Identification Method From Corn Based On ABC Algorithm And Probabilistic Neural Network

It is really important for speeding up the agricultural development to improve productivity and to upgrade the system. While weeds grow and reproduce in the fields, they compete with crops for water, light, soil nutrients and space, they disrupt agricultural production and the quality of the agricultural products and hinder the agricultural development.. Therefore, using weed control strategy to maintain the growth of crops is of crucial importance. At present, the weed control strategies include manual removing weeds, mechanical weeding, or the use of herbicides. Among them, the most economical and efficient way to clear the weeds in corn field is chemical control, but abuse and excessive of herbicides spraying can not only make resistant weeds but also cause environmental pollution and even make the herbicide residues enter the human body. Machine vision system can distinguish between crops and weeds, it can identify the appropriate weed area and spray through an automated system in different parts of the farmland acquisition images, then effectively applying herbicide to improve profitability and reduce environmental pollution. On the basis of summarizing the relevant research both at home and abroad, with a view to the defects of traditional weed identification algorithms such as low recognizing precision, and the poor real-time performance, this paper puts forward a kind of corn and weed identification method based on ABC and PPN, the main contents are as follows:(1) Collect a large number of corn weeds images and do super green method with image to enhance the green features. Encoding images prepare for the subsequent image processing algorithm. The images of acquisition are mainly about green plants and soil background, then analyzing the images the color of the RGB color space model component, choosing the appropriate characteristics, processing the three components of color image to enhance its green feature and getting plant prominent gray image.(2) Using the two-dimensional OTSU divides the original image into two parts of the foreground and background quickly. According to the characteristics of the weed image and the real-time demand of recognition, use the OTSU segmentation algorithm to get the binary images of the above-treated plants.(3) Mathematical morphology can process the image shape details, it has extensive and successful application of precede in terms of image detection, machine vision. For the wild point of segmented binary image, this paper adopte the morphology processing algorithm of open operation and close operation to make the image features stand out.(4) Researching how to extract the characteristics of corn and weed leaves: first, through analysis and calculation of the plants pixel points of each image to extract the six color characteristics of the plant. Second, extract the seven moment invariant features,invariant moment mainly embodies the geometric feature of image region, it has the invariant features of translation, scale and rotation.(5) Research to extract the optimal combination of characteristic method and construct weeds and corn classifier to identify weeds: corn and weed identification was proposed based on ABC algorithm and probabilistic neural network. Using the ABC algorithm to extract the optimal feature combination of corn and weed can improve the defects of a large number of characteristics redundant input laborious. Applying the ABC algorithm iterative selecting optimal adaptive PNN’s smoothing factor to improve the recognition performance of PNN. Finally, using the optimized characteristics vector as input of PNN, using the weeds or corn binary data as the output of training neural networks, thereby identifying its belong to corn or weeds.(6) Applying ABC algorithm selects optimal smoothing factor of PNN.The traditional method to select PNN smoothing factor of the selected parameters is experience value. In this way, we cannot select the most effective parameters used for digital image of corn and the classification of weeds according to different training test set adaptively, but increasing the ABC algorithm of PNN weed identification makes up for the shortage, and the choice of features to improves the efficiency of classification, so relative to the traditional PNN and SVM weed identification algorithm, the PNN weed identification algorithms combined with the ABC algorithm significantly improve recognition accuracy and recognition speed.