Article Preview
Top1 Introduction
Citrus fruits are one of the most popular fruits in the world, and they are also an important part of the agricultural economy. However, at the beginning of the 20th century, citrus huanglongbing was found in many areas of southern China for the first time. Later, citrus huanglongbing was discovered in other provinces (Sichuan, Guangxi and Jiangxi) in citrus production areas. The pathogen of Huanglongbing is a bacterial pathogen that is mainly transmitted by psyllids (Grafton-Cardwell et al., 2013), and it is one of the biggest virus hazards affecting citrus cultivation in the world. The pathogen is extremely infectious and has a very high pathogenicity rate. Once a citrus tree is infected with Huanglongbing pathogen, the citrus tree will lose the ability to grow fruits in a short time, which will cause the production of citrus to drop sharply. Therefore, the citrus huanglongbing of citrus growers has evolved into a state of “talking about Huanglongbing discoloration”. This has led to the research on the identification and prediction of citrus Huanglongbing becoming an urgent problem to be solved. Scholars have conducted in-depth research on the characteristics of citrus Huanglongbing (Gottwald, 2010). From the current development analysis of citrus Huanglongbing detection and identification technology, near-infrared spectroscopy technology and detection and prevention technology based on the Internet of Things have become the mainstream development of the technology for identifying citrus Huanglongbing. Near-infrared spectroscopy analysis technology has the advantages of rapid identification and better reliability of results (H. Xiao, 2019). The detection and prevention technology of citrus Huanglongbing based on the Internet of Things has the advantages of low cost and rapid identification, but its recognition rate needs to be improved. The key to improving its recognition rate is to find an effective disease image recognition algorithm. So, citrus Huanglongbing has attracted the attention of domestic and foreign scholars, and they have published many research results. Sankaran et al. used a model based on deep convolutional networks to identify citrus Huanglongbing. The authors studied whether periodic or “temporary” models are effective for tasks involving sequence, vision and other aspects, and developed a cyclic convolution architecture. This is suitable for end-to-end trainable large-scale visual learning. Authors demonstrate the value of these models in benchmark video recognition tasks, image description and retrieval problems, and video narration (Sankaran et al., 2013). Mishra et al. study the potential ability of visible light and near infrared (VIS-NIR) spectroscopy to identify Huanglongbing-infected citrus trees (Mishra et al., 2012). Two different spectroradiometers with a spectral range of 350 to 2500 nm are used to collect the canopy reflectance spectrum data, and three classification techniques are used to classify the data: nearest neighbor (KNN), logistic regression (LR) and support vector machine (SVM). The authors concluded that using only one canopy reflectance observation per tree is not enough. Due to the large differences in canopy reflectance data, no classification method can successfully distinguish healthy trees from Huanglongbing infected trees. Alireza et al. extracted several types of texture features from leaf images, and used five different feature selection methods to rank the best feature set that can describe infection features. The performance of the seven classifiers is based on a stepwise classification method. An assessment was conducted (Alireza, 2013). However, using this method will incorrectly classify some undernourished samples into other categories. In view of the similarity of characteristics between the characteristics of the disease images and the large dimension, especially the characteristics of the disease images may also be related, resulting in the generation of feature redundancy, which has a serious impact on the efficiency and accuracy of citrus Huanglongbing e image recognition. In addition, they have the defects of high cost of detection algorithms and low detection accuracy. This effect will occur in the image cutting feature extraction stage, so this paper uses the citrus Huanglongbing recognition algorithm based on CKMOPSO, which uses CKMOPSO to automatically select feature vectors with strong classification capabilities from the original disease image features for use in Huanglongbing image recognition, experimental results show that this is an effective recognition method.