Development of a vision system to evaluate greenhouse crop condition

Abstract

A modular software system designed to quantify the condition of greenhouse crops using computer vision and functional engineering methods is discussed. The platform integrates a full processing cycle which includes image acquisition, noise adaptation preprocessing, HSV-based vegetation segmentation, spectral and textural feature extraction, and analytical aggregation of phenotypic indices. The experimental evaluation was made on a heterogeneous data set consisting of 312 images with a resolution from 1280 × 720 to 4096 × 2160 pixels obtained under conditions of variable illumination and interference from sensors. The segmentation algorithm demonstrated high reliability maintaining contour integrity with brightness disturbances of ±40% and Gaussian noise levels up to σ = 20. Quantitative measurements, i.e. the ratio of plant areas, color fractions, the excess greenery index (ExG), and contrast and uniformity obtained on the basis of GLCM showed high sensitivity to physiological changes in chlorophyll content, pigmentation shifts, and structural heterogeneity of leaf tissue. The integrated health index, normalized to the interval [0, 1], effectively distinguished between non-stressed vegetation and stressed vegetation. Batch processing experiments confirmed the reproducibility of the calculated descriptors and the scalability of the system for high volume phenotyping tasks. The results demonstrate that the proposed solution provides a computational framework for real-time monitoring, early detection of vegetation stress, and integration into a digital agriculture decision support environment.