Citrus greening disease recognition algorithm based on classification network using TRL-GAN

https://doi.org/10.1016/j.compag.2022.107206Get rights and content

Highlights

  • Expanded the insufficient citrus greening diseased leaves to improve the generalization of the classification network. The diseased leaves with insufficient are expanded to strengthen the generalization of the classification network. Meanwhile, the generated citrus yellow dragon disease leaf fake data has the phenotypic characteristics of the real data, and the yellow dragon disease researchers are unable to identify the generated data real and fake, testing out TRL-GAN to achieve certain generation effect.

  • Effective background culling of direct collection of diseased leaf data in the field to achieve shooting effect under indoor conditions.

  • To make the classification network achieve a certain improvement in the accuracy rate of citrus greening diseased leaves.

Abstract

The monitoring and prevention and control of citrus yellow dragon disease is a significant measure to ensure citrus production. If yellow dragon disease appears in citrus orchards, it will cause root rot, fruit deformation and wilting of fruit trees, which will eventually spread to every fruit tree in the whole orchard and cause the death of fruit trees, so it is very meaningful to detect the symptoms of citrus yellow dragon disease early and take appropriate treatment and prevention measures. Pratically, the orchard owner will remove the corresponding fruit trees as soon as they are found to be infected with Huanglong disease, so that it is extremely problematic to obtain a large number of Huanglong disease leaf data. Meanwhile, due to the uncertainty of the pathological trait distribution of citrus yellow dragon disease leaves and the extreme shortage of data, the convolutional neural network model learned in a small number of samples is not capable of generalization. In order to improve the accuracy and generalization of Citrus Greening Disease recognition algorithm, this paper introduces Texture Reconstruction Loss CycleGAN(TRL-GAN) to generate citrus diseased leaf data in realistic scene to increase the richness of samples, and thus proposes the Recognizing Citrus Greening Based on TRL-GAN(RCG TRL-GAN). This algorithm firstly performs background culling by using the instance segmentation network Mask RCNN for realistic scenes citrus yellow dragon disease mottled, zinc deficiency, magnesium deficiency, leaf veins yellowing and other corresponding symptomatic leaves, then introduces texture reconstruction loss improvement CycleGAN as training and migrates the diseased leaf style to ordinary green leaves for data expansion, and finally uses the expanded dataset to train the convolutional neural network. Experimental results on the constructed dataset of 4516 images (762 mottled, 749 Zn deficient, 737 Mg deficient, 721 Vein yellowing, 783 Diachyma yellowing, 764 green leaves) reveal that TRL-GAN has 13.49% and 1.1% improvement in FID and KID, respectively, relative to the original structure CycleGAN, and has been identified by six citrus yellow dragon disease experts and three vision professionals identify that the fake data generated by TRL-GAN have similarity with the leaf pathological characteristics and real data, and also by using T-SNE technique it is observed that the real data have similar distribution with the generated fake data in two-dimensional plane. Meanwhile, the more outstanding accuracy performance in the classification network is ResNeXt101 with 97.45% accuracy, and the average accuracy of RCG TRL-GAN technique in the recognition of classification network is improved 2.76%. The study proves that the RCG TRL-GAN effectively improves the citrus greening disease phenotype data generation and recognition, and can provide method reference for the expansion and recognition of complex plant disease phenotype images.

Introduction

Citrus Greening Disease, known as citrus cancer, has caused serious damage to the citrus industry. It is extremely easy to spread by psylla - the disease’s infection source.Up to now,the most effective way to avoid the spread of citrus greening disease is to dig, burn and sterilize the diseased plants in time. The traditional detection methods for yellow dragon disease include field diagnosis, indicator crop identification, microscopic observation of the pathogen, serological identification, hybridization standards, DNA probe hybridization and PCR techniques. At present, the more accurate and reliable method is the PCR technique, which is time-consuming and costly. For field diagnosis, It is possible to be misdiagnosed by personal experience or diagnosis by phytopathologists (Deng et al., 2016). Citrus greening disease is diagnosed by phytopathologists mainly based on the appearance of a large number of mottled leaves (Garnier and Bové, 1993, do Carmo Teixeira et al., 2005, Varma et al., 1993, Teixeira et al., 2008). The mottled leaves display asymmetrical yellowing spots, which are distinct from the uniform yellowing caused by elemental deficiency, due to the gradual accumulation of Huanglong disease bacteria in citrus leaves caused by the necrosis of the leaf vein bast. Usually, untrained personnel can easily be confused with various types of citrus leaf yellowing when identifying mottled leaves. Therefore, at this stage, there is a need for a rapid and intelligent method to detect citrus yellow dragon disease to replace manual diagnosis.

In recent years, there have been a large number of experiments using machine learning to identify diseased leaves instead of personal experience or specialists. Traditional plant disease leaf recognition mainly extracts features through the corresponding filters designed for color, shape, texture, etc. and uses these features to classify them through machine learning methods such as random forests, support vector machines, neural networks, and others (Jiang et al., 2020, Manavalan, 2020, Yadav et al., 2019, Sun et al., 2019, Bischoff et al., 2021). There are some identification methods of citrus greening disease. For example, Xiaoling Deng et al. (2016) employed the C-SVC method to detect Huanglong diseased leaves, and Gómez-Flores et al. (2019) extracted features by producing diseased leaf data based on constant light brightness and then conducted random forest classification. These methods can solve the majority of the problems, but they involve human feature selection and cost a huge amount of time.

Because of the limited number of manually selected features, and the wide variety of crop diseases, manual extraction of features requires a lot of labor and time costs, resulting in the accuracy of traditional machine learning recognition is hard to achieve practical application standards. Nowadays, deep convolutional models can be used to learn features instead of manual feature extraction. With deep learning technology and increasingly sufficient plant disease image data, the traditional machine learning shows that the problems of extracting plant features and poor generalization ability are effectively solved, and many plant diseases can be recognized simultaneously by using deep convolutional networks instead of traditional manual features extraction (Chaudhary et al., 2020, Ma et al., 2018, Li et al., 2020, Zhou et al., 2021, Waheed et al., 2020, Zeng and Li, 2020, Zhang et al., 2019b, Zhang et al., 2019a, Zhang et al., 2021, Gao et al., 2021, Wang et al., 2021, Abade et al., 2021, Jiang et al., 2021, Zhong and Zhao, 2020, Zhang et al., 2019b, Zhang et al., 2019a, Gui et al., 2021). Likewise, deep learning has been made a certain achievement in the application of citrus greening disease at this point. For instance, Utpal Barman et al. (2020) applied MobileNet to recognize citrus diseased leaves, which include citrus green leaves, yellow dragon disease diseased leaves, and fast fading diseased leaves.However, once the plant is infected with the virus, it takes a long time for relevant phenotypic characteristics to emerge in reality. Citrus orchards are cleaned up as soon as the symptoms of the citrus greening disease appear, making data scarce and more difficult to collect and produce. Consequently, the amount of disease leaf data fails to meet the needs of the recognition algorithm data training of citrus greening disease.

To address the above problem of limited samples of plant diseases, Zhong et al. (2020) used a CVAE-like structure CAAE for citrus disease leaf data expansion, and used Zero- and few-shot learning for recognition based on these data. The approach can tackle the shortage of sample quantity and also gets effective results with a novel few-sample learning method for classification. However, the VAE series network is a display generative network structure, and there exists a more blurred generated image and a more concentrated distribution of generated data.

Since Zn-deficient leaves, Mg-deficient leaves, yellow veins leaves, and yellow diachyma leaves are easily confused with mottled leaves, this paper focuses on the phenotypes of mottled leaves, Zn-deficient leaves, Mg-deficient leaves, yellow veins leaves, and yellow diachyma leaves. This paper also improves CycleGAN, and takes advantage of the fact that it does not require a large amount of manual annotation, and thus proposes a Texture Reconstruction Loss CycleGAN (TRL-GAN) to expand the number of citrus yellow dragon disease leaves in realistic scenes and improve the quality of the generated images. In the following, the preparation of the dataset will be described, then the principle of the algorithm in this paper will be introduced, and finally the analytical experiments will be conducted to analyze the presence of techniques such as T-SNE in the experiments (Van der Maaten and Hinton, 2008).

Section snippets

Dataset collection

The data of this experiment mainly came from the public dataset PlantVillage and the photo collection of citrus diseased plants from the College of Plant Protection, South China Agricultural University. The data collection scene is shown in Fig. 1.

The device for collecting data of the citrus diseased plants in College of Plant Protection of South China Agricultural University was a OnePlus 8 T phone with a 48MP camera, 15 cm–30 cm shooting distance from diseased leaves, 35 mm equivalent focal

Results and discussion

To analyse the TRL-GAN results, this paper validates the effectiveness of the TRL-GAN network by using generative adversarial network evaluation indexes, subjective evaluation indexes, and classification networks.

Conclusions and outlooks

This paper mainly focuses on TRL-GAN for generating phenotype data of citrus greening disease, and to improve the performance of classification networks for citrus diseased leaf identification. Compared to the small dataset, this paper expands the dataset by Mask-RCNN to cull the background of the photographed leaves, followed by image synthesis using a more reasonable non-pairwise GAN technique. What’s more, in addition to introducing the original loss function of CycleGAN, it also adds

CRediT authorship contribution statement

Deqin Xiao: Conceptualization, Supervision, Investigation, Funding acquisition. Ruilin Zeng: Conceptualization, Methodology, Software, Validation, Investigation, Formal analysis, Data curation, Writing – original draft, Writing – review & editing. Youfu Liu: Conceptualization, Methodology, Data curation. Yigui Huang: Data curation. Junbing Liu: Data curation. Jianzhao Feng: Data curation. Xinglong Zhang: Supervision.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

This work was supported by the Science and Technology Planning Project of Guangzhou (Air, space and ground integrated intelligent planting monitoring and pest early warning system, grant number 202206010116), and Agricultural Research Project and Agricultural Technology Promotion Project of Guangdong (grant number 2021KJ383).

References (38)

Cited by (11)

View all citing articles on Scopus
View full text