A rice field weed identification method based on GA-ANN feature dimension reduction and SOM feature optimization

Abstract

The invention discloses a rice field weed identification method based on GA-ANN feature dimension reduction and SOM feature optimization. The method comprises the following steps: step 1, image acquisition; Collecting images of rice field weeds as samples under different natural conditions; Step 2, image preprocessing; In order to solve the problem that the weed identification precision is influenced by paddy field illumination intensity change and water surface reflection, on the basis of an RGB color feature segmentation algorithm, introducing an RGB linear weighting coefficient to segment an image to obtain a binary image, and then using a morphological operator for post-processing to separate out a weed binary image; According to the method, data redundancy can be reduced, the calculation amount is simplified, the screened main characteristics have the characteristics of independence, distinguishability and small number, the method can be effectively applied to weed classificationand recognition, and the classification precision is improved.

Description

technical field [0001] The invention relates to the field of weed identification in paddy fields, in particular to a method for identifying weeds in paddy fields based on GA-ANN feature dimensionality reduction and SOM feature optimization. Background technique [0002] Rice is one of the main food crops in my country, and weeds in rice fields are an important factor affecting the yield and quality of rice. Effective identification and accurate classification of weeds is a crucial step in realizing intelligent and precise weeding. Due to the characteristics of paddy fields, there are problems such as changes in light intensity and reflections on the water surface that have a great impact on the accuracy of weed recognition. In order to improve the accuracy of different weed recognition, it is necessary to analyze various characteristics of weeds (shape features, texture features, color features) Fusion application, but with the increase of the feature dimension, the recogni...

AI Technical Summary

Problems solved by technology

[0003] According to the literature review, most recognition algorithms are artificially set thresholds to process images, which have limited adaptability to light and cannot effectively reduce the impact of light intensity; most of them use feature fusion for dimensionality reduction, which reduces the complexity of classification and recognition, but it is difficult to analyze The significance of the influence of feature parameters on different types of weeds; for example, Li Xianfeng et al. proposed a multi-feature fusion weed identification method based on SVM and D-S evidence theory, which solved the problem of feature space complexity and high classification dimension. However, the recognition speed needs to be improved; Li Hui et al. used principal component analysis to reduce the dimensionality to obtain three principal components, which effectively reduced the training time of SVM and improved the classification accuracy, but the adaptability of the model was limited; Zhang Xinming et al. The problem is through the network's corn and weed recognition algorithm, which realizes the optimal feature selection and improves the classification accuracy, but cannot filter out highly correlated features, and the recognition accuracy needs to be improved

Images