Image Classification: Practical

A Brief Introduction to the Problem and the Task Addressed

The study area for pistachio image classification is located at the UNDER SUN Company farm in Armenia. This section introduces the study area, along with the available information and datasets.

The Armenian plateau is an important centre of cultivated plant origin, with a comparatively large number of crop wild relatives, numerous varieties of different species occurring in the country, as well as endemic species. Among the diverse plants of economic importance, nut crops hold a special role due to their unique and interspecific variability in their wild forms. The only wild species of Pistachio reported in the literature and grown in Armenia is Pistacia mutica Fisch. & Mey (Pistacia atlantica Desf. subsp. mutica (Fisch. et C.A.Mey.) Rech.f. (Anacardiaceae)). The scientific explorations attest to the cultivation of pistachios in Armenia's territory dating back to ancient times. Results of geo-mineralogical research of cuts of the boreholes put in Masriksky plain, and natural exposures of Sevan intermountain hollow have revealed prints of leaves and remains of pistachio plants (along with remains and prints of such plants as liana, juniper, oak, beech, pine) that allow to date them as Upper Sarmatian.

In various parts of Armenia, pistachio groves were established in the 1970s, which had significant soil protection benefits. The established areas are still preserved, despite their neglected condition. The "Under Sun" company restores a part of these areas, creating nurseries from various pistachio varieties.

In recent years, industrial pistachio orchards have been established in Armenia in the following regions: the Syunik region (10 ha), the Ararat region (20 ha), and the Armavir region (approximately 1000 ha). As well as small gardens in the Tavush and Kotayk regions.

"UNDER SUN" company started the establishment of a pistachio orchard in 2021 on an area of 150 hectares. Currently, the number of planted trees exceeds 60,000. "Pistachio natural" or "Vera" was chosen as the rootstock, which is maximally adapted to the local climatic conditions. This is evidenced by the previously planted almost 50-year-old pistachio soil protection groves. Currently, the trees are in the phase of continuous grafting. The orchard is based on a 3.5 x 6 m scheme with a 10% pollinator distribution. The first harvest is expected in 2025-2026.

According to soil types, the region includes sandy, sandy-clayey, clayey-sandy and clayey soils. Previously, the lands were not cultivated, so deep reclamation works were carried out.

According to the chemical composition of the soil, they are characterised by:

• With a very low content of organic matter, humus, not exceeding 0.8%.
• With a low content of nutrients, except for K, which in some cases reaches the minimum required level.
• With a high pH reaching up to 9.
• With a high content of dissolved salts (EC).

Pistachio orchard is prone to some diseases, which are:

• Septoria - expressed by drying of leaves and early leaf fall.
• Gray spotting - leaf spotting occurs.
• Curling - causes yellow-brown spots on the leaves.
• Powdery mildew - produces a white cotton-like coating.
• Root rot - manifested by a white or grey covering of the roots, especially in the root nodule, and dead cells.
• Verticillium Wilt is expressed by partial or complete drying of the plant. In the cross-section of the dried branches, there are strongly pronounced black marks.
• Blossom blight -expressed by the drying of young branches and leaves.
• Bacterial disease expressed by the drying of young branches, leaves and other parts.

Considering the mentioned diseases, preventive measures are planned during the entire vegetation period, which includes root and ecto-root treatments. Increasing the natural immune system of plants with proper nutrition and fertilisation is also a preventive measure.

For this purpose, considering the environment in which the plant grows and develops, fertilisation and nutrition processes are carried out using organic substances, such as humic acids, fulvic acids, and amino acids, as well as mineral fertilisers and microelements in the form of chelated compounds.

As a result of proper feeding and fertilisation, the plants grow intensively, allowing the garden to be harvested 1-2 years earlier.

Monitoring is also carried out in the garden, by which the following care activities are planned and implemented:

• Irrigation - according to the phonological stage of the tree, soil type, and climatic conditions.
• Weed control is carried out mechanically and chemically using herbicides.
• Mechanical works of the land - loosening, levelling.
• Trimming and shaping - as needed.
• With this background information, a key deliverable of this project is to prepare a map depicting the pistachio crop fields, which includes identification, classification, and area measurement using recent drone and satellite images for further analysis.

This case study aims to create an accurate crop map by utilising image classification techniques, drawing on data from Earth observation satellites, UAVs, and ground sensors, including images captured by cell phones.

Additionally, various data from the test area were stored in a GIS (including Google Earth). For example, information about soil, pistachio crop types or varieties. We also have polygon boundaries for each crop field and its corresponding crop type. In this case study, the available information may be used for:

• Training field selection for the classification of the satellite image
• Extraction of the prior knowledge
• Validation of the final results

Note: Any existing errors in the data and/or information used may impact the final results.

Conduct practical exercises: Selection of images, bands and preparation of data

• Define the information requirements.
• Mission planning: discuss mission planning to ensure quality data acquisition, processing and analysis. Address the limitations and complexities involved in operations by technicians.
• Access to a growing archive of satellite imagery and platforms, particularly Sentinel-2 data.
• Critical datasets selection for experimentation; the availability of new tools to facilitate image processing and interpretation, including improved EO data selection, visualisation, review, and analysis functions.
• Processing EO Satellite imagery, band selection, cloud masking, and image enhancement.
• QGIS exercises and Downloading Sentinel-2 satellite Imagery. The Copernicus Data Space Ecosystem Browser serves as a central hub for accessing, exploring and utilising the wealth of Earth observation and environmental data provided by the Copernicus Sentinel constellations. To download Sentinel-2 multispectral imagery, see the following tutorial from the Copernicus dataspace: https://documentation.dataspace.copernicus.eu/Applications/Browser.html

Conduct practical exercises in image classification using selected data

• Feature selection
• Remote sensing indices.
• Supervised image classification (Object classes and Training set selection, Maximum likelihood)
• Clustering and unsupervised image classification.
• QGIS exercises and case studies: Learn to create training samples, implement classification, assess classification accuracy, and apply these skills to practical case studies.

Generate independent, robust, and consistent reporting needs for image classification and estimate pistachio orchards/tree areas and their changes over time at regional, farm, and plot levels

• Accuracy assessment
• Generate error maps and evaluate differences
• Comparison of supervised and unsupervised results.

Collaborate with Pistachio growers to do knowledge-based classification

Measurements of the pistachio canopy's height, length, and width: Data collection should preferably be minimal and hypothesis-based. The hypotheses are supported by a method for collecting and processing the data, as well as feature extraction. The RGB, Depth cameras, or Drone would reduce the amount of work and improve the reliability of the required data.