Insect-related disasters are one of the most important factors affecting crop yield due to the fast reproduction and
widely distributed,
large variety of insects. In the agricultural revolution, detecting and recognizing insects plays an important role
in
the ability for
crops to grow healthily and produce a high-quality yield. To achieve this, insect recognition helps to differentiate
between bugs that
must be targeted for pest control and bugs that are essential for protecting farms. While the kinds of insects are
broad and available
insect datasets have been collected from different sources, the existing insect recognition models trained on a
specific dataset with particular,
predefined insects. Thus, the previous approaches on insect recognition are unable to recognize new types of insects
from other datasets as well
as utilize the knowledge from different datasets. Machine learning, especially deep learning requires a large volume
of data to achieve high performance
hence transferring and utilizing knowledge from various datasets are essential. Therefore, we propose a new deep
learning-based domain adaptation
algorithm to address these problems. By minimizing the gap of
distributions between different datasets, our proposed method can generalize well on the new target domains. In
addition, we design a hardware system so that we
deploy our deep learning model the complete system to run in real-world farms.
Our Solution
We introduce a completed hardware design for insect monitoring.
In particular, we propose a novel, end-to-end hardware that integrates a set of cameras, a motion detection
sensor,
a low-cost computer, and lighting modules to a unit system to capture and classify flying insects.
The system consists of two cameras that capture the top view and lateral view of insects. The motion sensor is
integrated to detect the
existence of the insects based on their movement into the imaging zone. Then, the sensor sends a signal to the
cameras to initiate image capture.
The system is designed to deploy in various conditions in which the environmental lighting is not ideal for taking
images (e.g., monitoring fields at night).
Hence, the LEDs are integrated to support and improve the quality of taken images. The cameras and LEDs have been
carefully timed by our hardware trigger
design to guarantee that the cameras take images at the same time as the LED flashes so that the taken images have
the best quality of brightness.
Finally, the taken images are sent immediately to the low-cost computer for further processing. Particularly, the
computer takes the images taken
from the two cameras and performs the AI algorithm to detect and classify the types of insects. The AI algorithm
has been developed
by our team by using the domain adaptation technique to train the AI model in an unsupervised manner.
Specifically, we develop the unsupervised domain adaptation technique to train the deep convolutional neural
network on the insect photos collected on the farms.
Also, the computational cost of the AI algorithm has been optimized so that we can effectively deploy on the
low-cost computer.
In addition, to attract the insects coming into the system, the described hardware and AI system will be attached
to an existing trap system that utilizes light
and semiochemicals to attract insects. To be able to deploy on the farm where the electric power is limited, the
power of the entire system is provided by solar energy.
Patents
[1]
Thanh-Dat Truong, Ashley PG Dowling, Randy J. Sasaka, and Khoa Luu.
Sensor-based Smart Insect Monitoring System in the Wild.
University of Arkansas Invention Disclosure, 2022.
[2]
Thanh-Dat Truong, Ashley PG Dowling, and Khoa Luu.
Smart Insect Control Device via Artificial Intelligence in Realtime Environment.
University of Arkansas Invention Disclosure, 2022.
Papers
[1] Thanh-Dat Truong, Chi Nhanh Duong, Pierce Helton, Ashley Dowling, Xin Li, and Khoa Luu.
CoMaL: Conditional Maximum Likelihood Approach to Self-supervised Domain Adaptation in Long-tail Semantic
Segmentation.
IEEE Transactions on Pattern Analysis and Machine Intelligence (Under Review).
[2] Thanh-Dat Truong, Ngan Le, Bhiksha Raj, Jackson Cothren, and Khoa Luu (2023). FREDOM: Fairness
Domain Adaptation
Approach to Semantic Scene Understanding. In Proceedings of the IEEE/CVF Conference on Computer Vision
and Pattern
Recognition (CVPR), 2023.
[3]
Pierce Helton, Ashley Dowling, and Khoa Luu.
Artificial Intelligence System for Automatic Imaging, Quantification, and Identification of Arthropods in
Leaf Litter and Pitfall Samples.
Inquiry Journal, 2022.
[4]
Thanh-Dat Truong, Naga Venkata Sai Raviteja Chappa, Xuan Bac Nguyen, Ngan Le, Ashley Dowling, and Khoa Luu.
OTAdapt: Optimal Transport-based Approach For Unsupervised Domain Adaptation.
In Proceedings of International Conference on Pattern Recognition (ICPR), 2022.
[5] Thanh-Dat Truong, Chi Nhan Duong, Ngan Le, Son Lam Phung, Chase Rainwater, and Khoa Luu.
BiMaL: Bijective Maximum Likelihood Approach to Domain Adaptation in Semantic Scene Segmentation.
In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021.
[6] Jalata, Ibsa, Naga Venkata Sai Raviteja Chappa, Thanh-Dat Truong, Pierce Helton, Chase Rainwater, and
Khoa Luu.
"EQAdap: Equipollent Domain Adaptation Approach to Image Deblurring." IEEE Access 10 (2022):
93203-93211.