Tomato Health and Fruit Count Check Using Smart Algorithm

Jim Crocker
18th April, 2024

Tomato Health and Fruit Count Check Using Smart Algorithm

This figure from the study illustrates the process of acquiring and preparing the image dataset, showcasing the autonomous mobile platform used for data collection in a tomato field (a), examples of the resulting labeled images (b), and the three annotated categories of unhealthy leaves, healthy leaves, and tomato fruits (c).

Image adapted from: Kang et al. / CC BY (Source)

Key Findings

  • In Jiangsu, a new system uses AI to detect tomato diseases and count fruits
  • The system's YOLO-TGI model identifies leaf diseases with high accuracy and low resource use
  • For counting tomatoes, YOLO-TGI-S paired with Byte-Track is fast and precise
In the realm of agriculture, particularly in the cultivation of tomatoes, farmers face the dual challenge of disease management and accurate fruit counting. Diseases can ravage crops, leading to significant yield losses, while manual fruit counting is laborious and imprecise. A recent study by the Jiangsu Academy of Agricultural Sciences[1] has made strides in addressing these issues through the development of an intelligent surveillance system that utilizes advanced deep learning techniques to monitor tomato plant health and fruit production. Tomatoes are susceptible to a variety of diseases and pests that can drastically affect their growth. Timely and effective control of these diseases is crucial for farmers to ensure high yields[2]. Traditional methods of disease and pest recognition involve complex steps such as image preprocessing and feature extraction, which can be time-consuming and inefficient. Deep learning, a subset of artificial intelligence, has revolutionized this process by enabling computers to learn from data directly, simplifying the recognition process[3]. One such deep learning approach, YOLO (You Only Look Once), has been particularly impactful in object detection tasks, including identifying diseases and pests in crops[2]. YOLO's ability to process images in real-time has significantly reduced the losses caused by these agricultural challenges. The study at hand builds upon this technology, enhancing it further to meet the specific needs of tomato cultivation. The researchers developed an improved version of the YOLO deep learning network, dubbed YOLO-TGI, by integrating Ghost and CBAM modules, which optimize the model for efficiency and accuracy. The Ghost module reduces computational requirements without sacrificing performance, while the CBAM module focuses the network's attention on relevant features in the image, improving its ability to recognize diseases. In parallel, another aspect of modern tomato cultivation has been addressed: the detection of tomato leaf diseases using a novel image-based approach called PLPNet[4]. This method effectively differentiates between diseases by focusing on distinctive characteristics and minimizing background interference, such as soil, which can be particularly problematic when diseases occur at the leaf's edge. Furthermore, the study incorporated state-of-the-art tracking algorithms to count fruits in video streams, a task that has traditionally been challenging due to the variability in the appearance of fruits and their environment. Deep learning has shown promise in this domain as well, with algorithms like MaskRCNN not only detecting objects but also the specific pixels belonging to each fruit, facilitating more accurate counting[5]. The integrated cascade framework proposed by the researchers synergizes the enhanced YOLO-TGI network with advanced trackers like Byte-Track, Motpy, and FairMot. This combination was found to be highly effective in both identifying tomato leaf diseases and counting fruits. The YOLO-TGI-N model, in particular, stood out for its low computational demands, achieving a mean average precision (mAP) of 0.72 for leaf disease detection with minimal resource usage. For fruit counting, the YOLO-TGI-S and Byte-Track pairing excelled, demonstrating a high correlation coefficient (R2) of 0.93 and a low root mean square error (RMSE) of 9.17, indicating precise counting with minimal error. Moreover, this combination was found to be significantly faster than other lightweight detection models, which is critical for real-time applications. The significance of this research lies not only in its immediate application to tomato cultivation but also in its broader implications for the agricultural sector. The framework developed by the Jiangsu Academy of Agricultural Sciences serves as a potential model for deploying similar surveillance systems across various types of fruit and vegetable crops, thus aiding in the modernization of agriculture and the push towards automation. This study not only ties together previous findings[2][4][5] but also expands upon them, offering a practical and scalable solution for farmers worldwide.

AgricultureBiotechPlant Science

References

Main Study

1) Toward Real Scenery: A Lightweight Tomato Growth Inspection Algorithm for Leaf Disease Detection and Fruit Counting.

Published 17th April, 2024

https://doi.org/10.34133/plantphenomics.0174

Related Studies

2) Tomato Diseases and Pests Detection Based on Improved Yolo V3 Convolutional Neural Network.

https://doi.org/10.3389/fpls.2020.00898

4) A Precise Image-Based Tomato Leaf Disease Detection Approach Using PLPNet.

https://doi.org/10.34133/plantphenomics.0042

5) Tomato Fruit Detection and Counting in Greenhouses Using Deep Learning.

https://doi.org/10.3389/fpls.2020.571299


Related Articles

An unhandled error has occurred. Reload 🗙