Plant diseases threaten global food security and cause significant financiallosses in agriculture. Early detection and precise diagnosis are critical for effective disease management. This study explores a novel approach to plant disease identification using the You Only Look Once (YOLOv12) algorithm combined with few-shot learning techniques. By leveraging a limited dataset, the model is trained to classify citrus plant leaf images into four categories: healthy, greening, black spot, and canker. The proposed system enhances disease detection efficiency, enabling farmers to take timely preventive measures. Our approach demonstrates the potential of few-shot learning in agricultural disease diagnosis, reducing the need for extensive labeled datasets while maintaining high accuracy.
Introduction
In India, major plant diseases such as rice blast, bacterial leaf blight, wheat rusts, citrus greening, and Panama disease severely impact staple crops, causing significant yield losses. Climate variability and soil persistence complicate management. New detection methods using molecular diagnostics (PCR, LAMP) and IoT-based digital monitoring are improving early disease identification, but accessibility remains a challenge for many farmers.
Traditional disease identification is difficult and labor-intensive, especially for large farms. Early detection, especially of leaf diseases, is critical for maintaining plant health.
An ideal modern plant disease diagnosis system would leverage few-shot learning (FSL) and computer vision (CV) to enable accurate, rapid identification of diseases from images captured by smartphones or drones, even with limited training data. Integrating environmental data would allow personalized management advice, reducing pesticide use and promoting sustainability. This technology could democratize expert knowledge, benefiting regions with limited agricultural support.
Key innovations include building extensive disease image databases through crowdsourcing, applying transfer and meta-learning for model adaptability, and using semi-supervised learning to improve accuracy with minimal labeled data. Real-time feedback would provide tailored treatment recommendations.
The methodology centers on a convolutional neural network trained on a large labeled source dataset and fine-tuned on limited target data using transfer learning, incorporating semi-supervised and iterative approaches to enhance performance by leveraging pseudo-labeled data selected with high confidence.
Experiments on various domain splits and few-shot learning setups show that semi-supervised methods, especially iterative semi-supervised learning, consistently outperform baseline transfer learning in accuracy, demonstrating the effectiveness of the approach in adapting to diverse crops and disease conditions.
Conclusion
Automatic classification of plant leaf diseases based on a few labeled samples is significant to guarantee the yield and quality with low cost of data. In this work, we proposed the semi-supervised few-shot learning scheme, which can improve the average accuracy of few-shot classification by adaptively selecting the pseudo-labeled samples to help fine-tune the model. Through literature research, to our best knowledge, we carried out the first semi-supervised work in the field of few-shot plant diseases classification. The PlantVillage dataset was divided into three split modes, and extensive comparison experiments were executed to prove the correctness and generalization of proposed methods. Considering all the different domain splits and k-shot, the average improvement by the proposed single semi-supervised method is 2.8%, and that by the iterative semi-supervised method is 4.6%. The study utilized the PlantVillage dataset, a publicly available collection of over 54,000 images of both healthy and diseased plant leaves across 38 different classes, representing 14 species of crops. These images were captured under controlled environmental conditions with uniform backgrounds. While widely used as a benchmark in plant disease classification , the controlled nature of the PlantVillage dataset might limit the generalizability of models trained on it to the more complex and variable conditions found in real-world agricultural fields.
References
[1] Liu J, Wang X. Plant diseases and pests detection based on deep learning:a review. Plant Methods. 2021;17(1):1–18.
[2] Wen BQ, Li Y, Kan Z, et al. Experimental study on microstructure andmechanical properties of stalk for Glycyrrhiza Glabra. J Biomech.2021;118:110198.
[3] Sheng X, Li Y, Lian M, et al. Influence of coupling interference onarrayed eddy current displacement measurement. Mater Eval.2016;74(12):1675–83.
[4] Nie J, Li Y, She S, et al. Magnetic shielding analysis for arrayed eddy current testing. J Magnet. 2019;24(2):328–32.
[5] Wang Y, Long A, Xiang L, et al. The verification of Jevons’ paradox of agricultural Water conservation in Tianshan District of China based on Waterfootprint. Agric Water Manag. 2020;239:106163.
[6] Nie J, Wang N, Wang K, et al. Effect of drip irrigation with magnetisedwater and fertiliser on cotton nutrient absorption. Earth Environ Sci.2021;697(1):012009.
[7] Li Y, Chao X. ANN-based continual classification in agriculture. Agriculture.2020;10(5):178.
[8] Garhwal AS, Pullanagari RR, Li M, et al. Hyperspectral imaging for identification of Zebra Chip disease in potatoes. Biosys Eng. 2020;197:306–17.
[9] Gao Z, Khot LR, Naidu RA, et al. Early detection of grapevine leafrolldisease in a red-berried wine grape cultivar using hyperspectral imaging.Comput Elect Agric. 2020;179:105807.
[10] Hu G, Yin C, Wan M, et al. Recognition of diseased Pinus trees in UAVimages using deep learning and AdaBoost classifier. Biosys Eng.2020;194:138–51.
[11] Misra T, Arora A, Marwaha S, et al. SpikeSegNet-a deep learning approachutilizing encoder-decoder network with hourglass for spike segmentation and counting in wheat plant from visual imaging. Plant Methods.2020;16(1):1–20.
[12] Liu J, Wang X. Early recognition of tomato gray leaf spot disease based onMobileNetv2-YOLOv3 model. Plant Methods. 2020;16:1–16.
[13] Too E, Yujian L, Njuki S, et al. A comparative study of fine-tuning deeplearning models for plant disease identification. Comput Electron Agric.2019;161:272–9.
[14] Nagasubramanian K, Jones S, Singh AK, et al. Plant disease identificationusing explainable 3D deep learning on hyperspectral images. Plant Methods. 2019;15(1):98.
[15] Jiang F, Lu Y, Chen Y, et al. Image recognition of four rice leaf diseasesbased on deep learning and support vector machine. ComputElectrAgric. 2020;179:105824.