Dataset of date palm tree thermal images and their classification based on Red Palm Weevil infestation

Adnan Nadeem Al Hassan^1*Muhammad Ashraf²Amir Mehmood³Kashif Rizwan²Muhammad Shoaib Siddiqui¹

• ¹Islamic University of Madinah, Medina, Saudi Arabia
• ²Federal Urdu University of Arts, Sciences and Technology Islamabad, Islamabad, Islamabad CT, Pakistan
• ³Department of Computer Science, Al-Kawthar University, Karachi 75300, Pakistan, Karachi, Punjab, Pakistan

The date palm trees are revered for their nutritious & high-energy fruits. They are drought-tolerant and thrive in hot, arid regions. Each tree has a single sturdy trunk covered with persistent leaf bases. The male tree produces pollen while the female tree bears fruit. The date palm trees (Phoenix dactylifera) are of significant economic and cultural importance in various regions worldwide, particularly in arid and semi-arid climates like those found in Khairpur, Sindh, Pakistan. However, these trees are susceptible to the Red Palm Weevil (RPW) pest, bayoud disease, and environmental stresses that adversely affect their productivity and overall health. Timely detection and management of these issues are crucial for sustainable date palm cultivation. The date palm is a nutritious and high-energy fruit consumed globally. According to statistics from Nationmaster (2020), Pakistan ranks as the sixth leading producer and consumer of dates worldwide. Remote sensing techniques, including RGB and thermal imaging, offer a non-invasive and efficient means of monitoring plant health and detecting stress factors. RGB imaging provides visual information about the foliage and appearance of date palm trees, while thermal imaging offers insights into their physiological status by measuring the trunk and crown temperature variations. Integrating these two modalities can enhance the accuracy and reliability of plant health assessment. El-Habbab et al. (2017) The decrease in production is primarily attributed to water scarcity and inappropriate pest management.The local farmers rely on manual methods for identifying infected trees through on-ground routine surveys, which are inefficient and sometimes delayed, allowing pests to spread to nearby trees. This late detection causes the need for timely identification and intervention to prevent further damage. This study emphasizes the critical importance of early detection to mitigate the spread of pests and preserve healthy trees. Therefore, decision-making requires a comprehensive and dependable dataset.Table 1 shows that Hamaidi (2019) published the dataset named "Date Palm data" which encompasses a collection of 2631 images capturing various aspects of date palm health, including healthy leaves, brown spots disease, and white scale infection which contain 1203, 470 and 958 files, respectively. In contrast, Al-Saad et al. (2022) Palm-Tree-Dataset focuses on remote sensing data tailored for palm tree detection, primarily for research and educational purposes, with specific emphasis on Al Ain Municipality in the UAE.Similarly, "Image dataset of infected date palm leaves by dubas insects" in Al-Mahmood et al. (2023) offers digital images of infected and uninfected palm fronds. The palm fronds are categorized as: 1) healthy palm leaves; 2) leaves infected by Dubas bug; 3) leaves infected by honeydew; 4) leaves infected by both Dubas bug and honeydew. The images of the leaves that the insects infect include different stages of an insect's life cycle, ranging from the third generation of nymphs to the adult stage in the fifth stage of the nymph. All the images are collected from a distance of 0.5 to 2 meters. The Dataset of infected date palm leaves for palm tree disease detection and classification Namoun et al. (2024) Type of data RGB and thermal images of date palm trees captured from three fields of view (FOVs). Each image in the dataset is annotated with farmers' feedback regarding the health status of the corresponding tree, categorizing it as non-infected, infected, badly damaged, or dead. How data is collected?Data was collected through on-ground surveys at various temporal setups (including dawn, morning, late morning, afternoon, and evening) during the summer season of 2023. A SeekShot IP54 thermal imaging camera with SEEKFUSION was used to capture the RGB color and thermal images of the date palm trees. The camera has a high-performance 206 × 156 thermal sensor; however, it combines this thermal sensor with a 640 × 480 color sensor. The SEEKFUSION technology generates 640 × 480 RGB and thermal images with precise detail and accuracy. As concealed activities of RPW lie in the lower and upper parts of the trunk, temperatures were captured from three fields of view (FOVs). All the trees were categorized as non-infected, infected, badly damaged, or dead based on visual inspection and farmers' feedback. Later on, this classification was validated by applying the histogram-based thermal analysis.The camera's resolution, thermal sensitivity (NETD), field of view (FOV), and emissivity of the targeted tree, as well as external factors like ambient temperature, distance, and viewing angle.The temperature of a subject tree in Celsius, its visual appearance, clock time, and farmers' feedback to classify among non-infected, infected, badly damaged, and dead. This article presents a comprehensive dataset for controlling and monitoring the RPW infestation. The dataset aims to facilitate: 1) RPW pest management; 2) the development and evaluation of machine learning algorithms for automated date palm tree classification by providing a diverse collection of images and ground truth labels. It contains 832 images (RGB and thermal) collected through the on-ground survey in date palm fields. Each image in the dataset is annotated with farmers' feedback regarding the health status of the corresponding tree, categorizing it as non-infected, infected, badly damaged, or dead. A neural network-based thermal analysis is made to validate the initial classification; further investigation of this dataset is left for users as future endeavors. The dataset shows promising potential for further studies, supporting a wide range of research topics extending the body of knowledge. It is valuable for researchers, industry professionals, public authorities, and others interested in harvesting date palm trees.The dataset was particularly developed for RPW pest management. All the thermal and color images were collected from 179 trees, which we selected from various date palm groves in Khairpur, Sindh, Pakistan.These trees were categorized into four classes: Non-infected (80 trees), Infected (32 trees), badly damaged (12 trees), and Dead (55 trees) based on visual inspection and the farmers' feedback in the field. We could find these trees with the help of a local navigator, who contacted local farmers and then led us to the various date palm groves for data collection. Some infected trees were deep inside the groves, while others were on the borders. Some of the infected trees were at the initial stages of the RPW infestation, and their physical appearance was quite normal due to the minimal damage. We also visited a date palm grove where all the trees were badly damaged. We found no overlap of trees in any date palm grove. Local farmers there don't use technology, instead, they employ human labour to manage their various tasks in the field manually.The dataset contains 832 processed images (RGB and thermal) of the Aseel cultivar, a predominant cultivar in Khairpur and Sukkur regions. Each image in the dataset is annotated with farmers' feedback regarding the health status of the corresponding tree, categorizing it as non-infected, infected, badly damaged, or dead. Table 2 briefly describes the specifications of the dataset. A few sample images are presented as an example in Figure 1, where the health status of the sample tree can be observed easily. This dataset serves as a valuable resource for researchers and practitioners working on precision agriculture and plant health monitoring by providing a diverse collection of images along with segregated folders.Future research endeavors can leverage this dataset to develop and evaluate machine learning algorithms for automated date palm tree classification and disease detection, ultimately contributing to the advancement of sustainable agriculture practices and food security in arid regions.Each tree in the dataset has a single sturdy trunk covered with persistent leaf bases, which form a thick coat that protects the bundles of vascular tissue in the tree trunk. These vascular bundles contain the tissues of Xylem and Phloem, where Xylem transports the minerals and water to the aerial part of the tree, and Phloem transports carbohydrates from the source to the sink. When a date palm tree is infested, the larvae inside the trunk gradually damage the vascular tissue bundles, which degrade the tree's efficacy for water utilization. This degradation leads the infected tree to a severe infection level where the recovery of that tree becomes impossible; hence, the tree finally dies. As the concealed damages gradually increase from an initial level to the final severe stage, internal fermentation raises the trunk's temperature at damage points.Therefore, temperature profiles of the trunk's surfaces can play a key role in pest management. We can use them to validate the initial classification based on visual inspection and the farmers' feedback.Thermal histograms can be used as the temperature profiles of thermal images; they are immune to the roughness of a thick coat on the trunk surfaces. The histograms of sampled images are presented in Figure 2, where a slight shift of the temperature envelope can be observed in temperature distributions.The distribution envelope slightly shifts towards the higher temperature if concealed fermentation rises due to the severity of internal damages. The temperature envelope of the dead and diseased trees shifts toward the right compared to the temperature envelope of the non-infected tree. Such information is very critical for conducting a comprehensive study on the health status of diseased date palm trees. The thermal analysis presented in section 3.2.3 also supports the concept that thermal analysis can provide insight into the concealed activities in tree trunks. Minimum and maximum temperatures of a thermal image are required to compute its thermal histogram. Therefore, these values are provided for every thermal image in the dataset. The process involved several key steps, including image acquisition, annotation, and validation; see Fig 3(a).RGB and thermal images were captured using a highly sensitive SeekShot IP54 thermal imaging camera, ensuring the comprehensive coverage of a date palm tree with three different fields of view (FOV), as shown in Fig 3(b). Every selected tree was carefully observed before image acquisition, then its health status was inquired about from the local farmers. After image acquisition, each image in the dataset was annotated with farmers' feedback regarding the health status of the corresponding tree, categorizing it as non-infected, infected, badly damaged, or dead. These annotations were carefully reviewed by the expert agronomists familiar with date palm cultivation and the domain knowledge. After raw data collection, all images of every tree were gathered in separate folders and then preprocessed and organized according to the hierarchy shown in Fig 3 (c). Every thermal image in processed data has a corresponding color image in RGB color images. The computation of a thermal histogram requires the image's minimum and maximum temperatures; therefore, these values are provided in a separate spreadsheet. The raw data is also presented in a separate folder.The analysis of thermal images mainly relies on thermal histograms, which cover all the parameters mentioned in Fig. 3(b). The histograms themselves or after further processing may construct unique feature