Abstract
In order to improve the utilization of the Agricultural Expert System so that it can be taken to the fields directly to guide agricultural production, the development and implementation of a portable Agriculture Expert System are studied in this paper. Specifically, image acquisition is proposed as the primary information input for human-computer interaction. By detailing the process of camera-based image capture, this research demonstrates an Agricultural Expert System that is easier to operate and can guide peasant production more directly.
Introduction
While agricultural expert systems have existed since the 1970s, they have traditionally been tethered to PC platforms. This creates a significant "last-mile" problem: farmers in rural areas—particularly in developing regions—often lack the technical infrastructure or the high-level literacy required to use complex, text-heavy software in the middle of a field.
The authors identify three main bottlenecks preventing widespread adoption: portability, interaction barriers, and hardware constraints. Desktop PCs cannot be easily carried to a diseased plant for real-time diagnosis, and most systems rely on text-based input that is inconvenient for field work. Furthermore, running large knowledge bases and processing images requires a balance of high performance and low power consumption that older embedded systems struggled to provide.
Hardware Architecture
The system is built on the Feiling S3C6410 development board, featuring an ARM11 processor. This chip was chosen because it offers a 40% performance increase over the older ARM9 architecture, supporting dual-instruction execution and efficient data handling. The platform includes 2GB of NAND Flash and 256MB of DDR memory to store the extensive knowledge bases required for diagnosis.
As shown in the hardware block diagram, the S3C6410 processor acts as the central hub, interfacing with a touch screen for user input, a TFTLCD for display, and USB ports for camera connectivity. This integrated approach allows for a compact, handheld form factor suitable for field use.
Software and Reasoning Engine
The core of the system is a data-driven "forward reasoning" engine. It starts with observed facts (the "antecedents"), matches them against rules in the knowledge base, and iterates until a conclusion (the diagnosis) is reached. The knowledge base for jujube (date) pest diagnosis was constructed using SQL Server and data mining techniques to extract patterns from historical agricultural data.
To solve the latency issues of transferring image data from a USB camera to the processor, the researchers implemented a memory-mapped (mmap) approach. This allows the image processing program to read the kernel-mode frame buffer directly, bypassing slow data-copying steps between the kernel and user space. This technical optimization ensures that the "portable" experience remains fluid even when handling high-resolution image data.
Key Results and Discussion
The transition to the S3C6410 (ARM11) platform provided a 40% increase in performance compared to previous ARM9-based systems, enabling the processing of complex diagnostic algorithms in the field. The integration of 2GB NAND Flash proved sufficient for storing large-scale agricultural databases and JPEG-compressed image files.
The authors successfully demonstrated a functional prototype capable of capturing images of diseased plants, processing them through a Linux-based software stack, and matching them against a knowledge base to provide prevention and treatment advice. By using images as a "universal language," the system bypasses the traditional bottleneck of text-heavy human-computer interaction.
Conclusion and Future Work
While the current system provides a robust foundation for portable diagnosis, the authors note that future work is required in image post-processing and automated analysis to improve diagnostic accuracy. Currently focused on jujube pests, expanding the system to a wider variety of crops will require significantly larger and more diverse knowledge bases. Future iterations aim to combine more advanced image recognition algorithms with the existing expert system logic to make the diagnosis fully autonomous.
Summary updated