Intelligent machine vision-based decision support system for lettuce growth monitoring

• Main Author: Loresco, Pocholo James M.
• Format: text
• Language: English
• Published: Animo Repository 2019
• Subjects: Computer vision; Decision support systems; Image processing; Machine learning; Lettuce; Computer Engineering
• Online Access: https://animorepository.dlsu.edu.ph/etd_doctoral/1502
• Institution: De La Salle University

Recently, computer vision approaches for plant growth monitoring are being increasingly employed. This made the plant growth monitoring continuous, automated and non-invasive. This kind of intelligent monitoring in green house farming allows precise understanding and analysis of various environmental factors attributed to growth and farm management deficiencies. This study presents a machine vision-based decision support system for determination of lettuce growth stage. Extensive computer vision descriptors such as color, morphological, and texture features classify growth of lettuce crops in 3 stages (sowing, vegetative, and harvest) using image processing techniques and machine learning. Color space analysis using K-Nearest Neighbor aided the selection of the color space from RGB, HSV, CIELab, and YCbCr appropriate for the smart farm setup. A scale invariant area calculation of lettuce leaf area was developed by detecting a template marker with a known area for normalizing area measurements. The template marker was automatically detected by Viola-Jones algorithm and knowledge-based methods using HOG, LBP and Haar-like features. Segmentation of lettuce plants used thresholding and Superpixels which classified lettuce plant and background pixels from images taken at a smart farm hydroponics setup. Lab color information of the image extracted from a training image dataset have undergone two-level thresholding and K- means clustering thru Superpixels to identify each pixel class. A methodology of classifying lettuce growth stage using a Hybrid Decision Tree-Fuzzy- Rough Set is also presented. Vision features were extracted and subjected to dimensionality reduction using Decision Tree. The reduced inputs were used to design the Mamdani Fuzzy Inference system. Rough Set Theory was then applied to the Fuzzy Logic model to simplify the rules. Further, this research also presented an ANN-based decision support system of classifying lettuce growth stage using extracted vision features that included 2 morphological features (area, perimeter), 12 color features (RGB, HSV, YCbCr, Lab) and 5 textural features (contrast, energy, correlation, entropy and homogeneity). Image processing techniques were used to extract the required vision features and the neural network was trained using scaled conjugate gradient back propagation. Based on the results obtained, CIELab color space is the best color space to be used in the identification of the growth stage of the lettuce. Area measurements produced by the system using a template marker performed well in terms of root-mean- square error (RMSE). Experimental testing results demonstrate an improved performance in segmentation using Superpixels and thresholding in terms sensitivity, precision, and F1-score. The decision support systems developed here exhibited promising results in lettuce growth stage classification.