Identification of nutrient deficiency stress for iron, zinc and manganese in baby spinach using computer vision

• A computer vision system was developed to detect deficiencies of micronutrients including iron, zinc, and manganese in spinach. • Varying levels of iron, manganese, and zinc deficiencies (from 0 to 100% of recommended amounts) were detected as soon as the true leaves of spinach emerged. • A relationship exists between the levels of iron, zinc, and manganese in spinach and the color, morphological, and texture parameters extracted from plant image. • ANN was more effective in identifying varying levels of micronutrient deficiencies in spinach compared to other classification algorithms such as SVM, KNN, Naive Bayes, and Random Forest. • The proposed classification model demonstrated higher accuracy than the expert in distinguishing four levels of iron, zinc, and manganese deficiencies in spinach during the cultivation period. • The proposed method enables farmers to address nutritional deficiencies in spinach plants early, preventing a decline in product quality. In recent years, various methodologies have been developed to evaluate the quality of agricultural products. These technologies have facilitated the monitoring of plant nutritional needs, thereby reducing reliance on human judgment and enhancing intuitive decision-making. This study focuses on the cultivation of baby spinach plants in soilless beds to precisely control nutrient conditions. The objective was to investigate deficiencies for three key micronutrients, namely, iron (Fe), zinc (Zn), and manganese (Mn), at five concentration levels, i.e., 0%, 25%, 50%, 75%, and 100% of the recommended amount. A set of color, texture and morphological features were extracted from the sample images. Feature selection was carried out using the ReliefF method and the Sequential selection method for four distinct modes: Sequential Forward Selection (SFS), Sequential Backward Selection (SBS), Sequential Floating Forward Selection (SFFS) and Sequential Floating Backward Selection (SFBS). Results showed that SFBS exhibited the best performance compared to the other algorithms. Subsequently, five classification methods (Artificial Neural Network (ANN), Naive Bayes (NB), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Random Forest (RF)) were evaluated using the selected features as input in a predetermined order based on the SFBS results. The best rankings were ANN, KNN, RF, SVM and NB, respectively. Hence, as soon as the true leaf of a baby spinach crop expands, the deficiency of three micronutrients can be detected by ANN at various levels with an accuracy of 88% for Fe, 84% for Mn and 76% for Zn. A hierarchical model was also applied, achieving over 70% accuracy.