Image Recognition of Rapeseed Pests Based on Random Forest Classifier

Image Recognition of Rapeseed Pests Based on Random Forest Classifier

Li Zhu (Hubei Collaborative Innovation Center for High-efficiency Utilization of Solar Energy, Hubei University of Technology, Wuhan, China), Minghu Wu (Hubei Collaborative Innovation Center for High-efficiency Utilization of Solar Energy, Hubei University of Technology, Wuhan, China), Xiangkui Wan (Hubei Collaborative Innovation Center for High-efficiency Utilization of Solar Energy, Hubei University of Technology, Wuhan, China), Nan Zhao (Hubei Collaborative Innovation Center for High-efficiency Utilization of Solar Energy, Hubei University of Technology, Wuhan, China) and Wei Xiong (Hubei Collaborative Innovation Center for High-efficiency Utilization of Solar Energy, Hubei University of Technology, Wuhan, China)
DOI: 10.4018/IJITWE.2017070101
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

Rapeseed pests will result in a rapeseed production reduction. The accurate identification of rapeseed pests is the foundation for the optimal opportunity for treatment and the use of pesticide pertinently. Manual recognition is labour-intensive and strong subjective. This paper propsed a image recognition method of rapeseed pests based on the color characteristics. The GrabCut algorithm is adopted to segment the foreground from the image of the pets. The noise with small area is filtered out. The benchmark images is obtained from the minimum enclosing rectangle of the rapeseed pests. Two types of color feature description of pests is adopt, one is the three order color moments of the normalized H/S channel; the other is the cross matching index calculated by the reverse projection of the color histogram. A multi-dimensional vector, which is used to train the random forest classifier, is extracted from the color feature of the benchmark image. The recognition results can be obtained by inputing the color features of the image to be detected to the random forest classifier and training.The experiment showed that the proposed method may identify five kinds of rapeseed accurately such as erythema, cabbage caterpillar, colaphellus bowringii baly, flea beetle and aphid with the recognition rate of 96%.
Article Preview

1. Introduction

The disaster from diseases and pests on plant is one of the three natural disasters in China. The research from the food and agriculture organization of the united nations shows that the loss rate only from the disaster of diseases and pests to natural is more than 37% (Cao, 2015). Rapeseed is primary oil crops in the world (Chen & Jian, 2010). Rapeseed pests is variety, area-widely and large harmly. It has restricted the improvement of the yield per unit and the quality. It is effective to protection of pest management and reduction of the pollution on environment that the pesticide can be selectively used for different kinds of the disaster from diseases and pests,which is the basic premise for the accurate judgement and the precision medicine (Pan et al., 2015).

At present, the main detection methods for rapeseed pest include manual detection, multispectral technology and computer vision, etc.. Manual detection methods rely on the naked eye to judge by the experience of agricultural workers. Labor intensity is large and it lacks of objectivity. Otherwise, the artificial classification is increasingly difficult because of the various kinds of insects (Zou, 2011). Although it has been studied widely in the laboratory environment for the characteristics of strong feature, high sensitivity, indestructible etc. multispectral detection is difficult to be directly applied in field environment under real-time online detection for it is easily affected by the environment (Qiao, Xia, Ma, Cheng, & Zhou, 2010; Yuan, Zhang, & Wang, 2012). The image sensors used by computer vision has the visual ability to distinguish far more than human’s. The high quality digital images are obtained and processed in accordance with expectation precisely and complexly, which is very applicable in online recognition (Macleod, Benfield, & Culverhouse, 2010; Yu & Dai, 2010).

The pest identification technology on computer vision can be divided into two categories: one is the reverse identification taking advantage of the characteristics by which is done harm to plant by pests. As proposed in the literature (Zhang, Wang, Xie, & Li, 2014), a crop pests image recognition based on multi-features fusion can accurately identify 34 kinds of pests of four crops, such as rice, rape, corn and soybean. In Zhang, Ji, Yuan, Li, & Qi (2011), the characteristics of color, shape and texture from the hurt cotton leaf were extracted A support vector machine (SVM) classifier with radial basis function were employed to classify the main rapeseed pests. The classification accuracy was 88. 1%. In Cai & He (2010), after image pre-processing of the eaten leaves by pests, seven spherical shape feature values of eaten leaves could be automatically extracted: roundness degree, complexity degree, and spherical degree, etc., and then the BP neural networkmodel for recognition could be built up. Multi-fractal analysis of Fourier transform spectra was adopted in Wen and Cao (2013) to investigate the possibility of extraction of damage pattern characteristics for Citrus reticulata Blanco var. Ponkan. A BP neural network model is established for citrus pest identification. A detection method to realise the fast detection of plant pests and diseases in agricultural production is proposed in Jiang, Lu, Feng, & Guo (2014). For four kinds of features of each plant leaf image,the colour,HSV, texture and directional gradient histogram, the method adopts the bag of features based on multi-features fusion approach to form the eigenvector,and uses SVM classifier to train the classification.

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 14: 4 Issues (2019): Forthcoming, Available for Pre-Order
Volume 13: 4 Issues (2018): 2 Released, 2 Forthcoming
Volume 12: 4 Issues (2017)
Volume 11: 4 Issues (2016)
Volume 10: 4 Issues (2015)
Volume 9: 4 Issues (2014)
Volume 8: 4 Issues (2013)
Volume 7: 4 Issues (2012)
Volume 6: 4 Issues (2011)
Volume 5: 4 Issues (2010)
Volume 4: 4 Issues (2009)
Volume 3: 4 Issues (2008)
Volume 2: 4 Issues (2007)
Volume 1: 4 Issues (2006)
View Complete Journal Contents Listing