Location: Fruit and Tree Nut Research
Project Number: 6042-21000-006-002-S
Project Type: Non-Assistance Cooperative Agreement
Start Date: Aug 1, 2022
End Date: Jul 31, 2024
Objective:
To develop computational technologies for data pre-processing, data segmentation, feature extraction, and classification using aerial images and other photos of peach fruit and trees, to facilitate collecting peach phenotype data. Specifically, the project will design image processing algorithms to process peach fruit and tree images to collect the phenotype data on fruit characteristics (e.g., fruit color, size, density, and distribution) and tree health status (healthy vs diseased) based on leaves, canopies and other distinguishable subjects.
Approach:
Typical fruit and tree images of different angles, known health statues, and known tree growth and fruit ripening stages will be taken at the SEFTNRL using an iPhone and a drone. These photos will be subjected to imaging process using algorithms detailed below. The project will primarily consist of two phases with 3 months for each phase. In phase I, the project will identify mature peach fruit and healthy peach trees of different angles based on designed image processing algorithms. In phase II, the project will detect peach tree fruit ripening stages (unripe vs. ripe - using blush color and fruit size as the primary indicator) and tree health statues (healthy vs. diseased - using phony peach disease as an example), based on modified algorithms in phase I.
Phase I: will focus on the implementation of image processing algorithms on the healthy peach tree including fruits and leaves. The possible application of phase I results is fruit grading.
1. Acquire the images from SEFTNRL: The SEFTNRL will capture the peach fruit and tree images. The images will be the input data of the research.
2. Pre-process images: Images will be removed noise, smoothened, and resized. If the noise covariance is known, Kalman filter will be applied. If the noise covariance is unknown, H infinity filter will be applied.
3. Image segmentation: An image will be divided into several segments. The main four categories of image segmentation are thresholding-based segments, edge-based segment, region-based segment, and energy-based segment.
4. Feature extraction: The features like color, texture, and shape will be obtained. The most used feature extraction is morphological features: area, perimeter, major and minor axis length, and aspect ratio. In the project, the peaches and leaves will be applied this feature algorithm to obtain the healthy status of peach tree.
5. Classification of the images: Based on results from step 4, the images will be organized based on the analysis of features. Algorithms such as minimum distance classifier or artificial neural networks classifier will be applied.
Phase II: Image processing algorithms will be used to identify diseases using brown rot and phony peach disease as examples. The former has symptoms primarily on fruit and the latter on tree canopy and other vegetative growth characteristics.
1. The images of peach fruit with brown rot and trees with phony peach disease. Those images will be the input data for research algorithms.
2. Perform step 2, step 3, and step 4 from phase I. In this procedure, the feature will be extracted.
3. Classification of the images: The peach fruit and trees with diseases will be detected and classified based on the classification algorithms.
