Potassium deficiency diagnosis method of apple leaves based on MLR-LDA-SVM

Front Plant Sci. 2023 Nov 29:14:1271933. doi: 10.3389/fpls.2023.1271933. eCollection 2023.

Abstract

Introduction: At present, machine learning and image processing technology are widely used in plant disease diagnosis. In order to address the challenges of subjectivity, cost, and timeliness associated with traditional methods of diagnosing potassium deficiency in apple tree leaves.

Methods: The study proposes a model that utilizes image processing technology and machine learning techniques to enhance the accuracy of detection during each growth period. Leaf images were collected at different growth stages and processed through denoising and segmentation. Color and shape features of the leaves were extracted and a multiple regression analysis model was used to screen for key features. Linear discriminant analysis was then employed to optimize the data and obtain the optimal shape and color feature factors of apple tree leaves during each growth period. Various machine-learning methods, including SVM, DT, and KNN, were used for the diagnosis of potassium deficiency.

Results: The MLR-LDA-SVM model was found to be the optimal model based on comprehensive evaluation indicators. Field experiments were conducted to verify the accuracy of the diagnostic model, achieving high diagnostic accuracy during different growth periods.

Discussion: The model can accurately diagnose whether potassium deficiency exists in apple tree leaves during each growth period. This provides theoretical guidance for intelligent and precise water and fertilizer management in orchards.

Keywords: apple leaves; diagnostic method; potassium deficiency; shape-color feature; support vector machine.

Grants and funding

The author(s) declare financial support was received for the research, authorship, and/or publication of this article. This research was funded by the China Agriculture Research System (CARS-27) and the Shandong Province Natural Science Foundation Youth Project (ZR2023QE237).