论文信息
论文解读
AI 辅助整理- 方法
- Python+Streamlit,MobileNetV2图像分类,随机森林作物预
- 发现
- 病害检测准确率约85–90%,作物预测约88–92%,但需更多田间数据验证。
- 机会点
- 多模块集成系统缺乏田间验证,可研究跨区域泛化与模型可解释性。
论文摘要
本文介绍了一款基于人工智能的应用,将多项常见的农业支持任务集成于一体。该系统融合了基于图像的植物病害识别、基于土壤与气象数值的作物预测、肥料建议、土壤肥力评估以及问答助手等功能。系统采用Python和Streamlit构建。植物图像分类使用基于MobileNetV2的卷积神经网络,而随机森林模型则根据氮、磷、钾、温度、湿度、pH值和降雨量数值预测适宜作物。肥料与土壤肥力输出通过预定义规则生成。由Ollama驱动的助手可对农业问题提供通用回答。项目报告记录病害检测的准确率约为85–90%,作物预测的准确率约为88–92%。这些数据为项目自报结果,仍需使用多样化的田间数据进一步测试。该应用旨在辅助农民决策,而非替代专家建议或实验室土壤分析。
全文
International Journal of Science Strategic Management and TechnologyInternational Journal of Science, Strategic Management and Technology
Volume 02 Issue 11 November-2026 | ISSN: 3108-1762 (Online) | Impact Factor: 3.8
An International, Peer-Reviewed, Open Access Scholarly Journal Indexed in recognized academic databases
© Author(s). This work is peer-reviewed, openly published, and permanently archived
This article is openly accessible and reusable with proper attribution.
https://ijsmt.org/ , Email: editor@ijsmt.org 1
AI Smart Agriculture System
Pradeep Gowda
Head of the Department
Dept. of CSE
Sri Revana Siddeshwara Institute of
Technology, Chikkajala, Banglore
562157
Ushashree B N
Assistant Professor
Dept. of CSE
Sri Revana Siddeshwara Institute of
Technology, Chikkajala, Banglore
562157
Ushashreebn@gmail.com
G Harshavardhan
Dept. of CSE
Sri Revana Siddeshwara Institute of
Technology, Chikkajala, Banglore
562157
gharshavardhangouda@gmail.com
Manu B
Dept. of
CSE
Sri Revana Siddeshwara Institute of
Technology, Chikkajala, Banglore
562157
14manu444@gmail.com
Raj Kumar AB
Dept. of CSE
Sri Revana Siddeshwara Institute of
Technology, Chikkajala, Banglore
562157
rajkumarballolli4622@gmail.com
Siddhu G
Dept. of CSE
Sri Revana Siddeshwara Institute of
Technology, Chikkajala, Banglore
562157
siddugowda760@gmail.com )
Abstract— This paper presents an AI -based application that
brings several common agricultural support tasks into one
place. The system combines plant disease identification from
images, crop prediction from soil and weather-related values,
fertilizer suggestions, soil fertility assessment and a question -
answering assistant. It is built with Python and Stream lit. A
MobileNetV2-based convolutional neural network is used for
plant-image classification, while a Random Forest model
predicts suitable crops from nitrogen, phosphorus, potassium,
temperature, humidity, pH and rainfall values. Fertilizer and
soil-fertility outputs are generated using predefined rules. An
Ollama-powered assistant provides general responses to
farming questions. The project report records approximate
accuracies of85–90% for disease detection and 88 –92% for
crop prediction. These figures are presented as project -
reported results and need further testing with varied field
data. The application is intended to support farmers’
decisions, not replace expert advice or laboratory soil analysis
.Index Terms — Smart agriculture, plant disease
detection,MobileNetV2, convolutional neural network,
Random Forest, crop prediction, fertilizer recommendation,
soil fertility, Stream lit..
I. INTRODUCTION
Background and Motivation:. Agriculture is affected by
many factors, including soil condition rainfall, temperature,
water availability, pests and plant diseases
.Farmers often have to consider several of these factors at the
same time, and useful information may be spread across
different tools or sources. This can make it difficult to identify
a problem early or compare possible crop choices.
Project Objective and Scope : Plant diseases can reduce
both crop yield and quality. A visual check may not always
be enough to distinguish diseases with similar symptoms.
Fertilizer decisions also depend on nutrient levels in the
soil; applying too much or too little can increase costs and
affect soil health. Crop selection, meanwhile, is
influenced by nutrients and conditions
Contribution: Recent developments in machine learning
and computer vision offer practical ways to analyse plant
images and agricultural data. This project brings several
such functions together in a single Stream lit application.
Its purpose is to provide accessible decision-support in
formation while leaving final farming decisions with the
user and qualified agricultural professionals.
.
II. PROBLEM DEFINITION & OBJECTIVES
Farmers may experience delays in identifying plant
diseases, uncertainty when choosing crops, and difficulty
deciding how much fertilizer to apply. These tasks are
often handled using separate tools. The project addresses
this issue by developing one application that combines
image-based disease detection, crop prediction, fertilizer
recommendations, soil -fertility analysis and general
agricultural guidance.
The main objectives are to support early identification of
possible plant diseases; suggest crops using soil and
environmental inputs ; provide fertilizer guidance based
on nutrient values; classify soil fertility from NPK
readings; and make general farming information easier to
access through a simple interface.
https://doi.org/10.55041/ijsmt.v2i11.024
Cite this Article: Gowda, P., N, U. B., Harshavardhan, G., B, M., AB, R. K. & G, S. (2026). AI Smart Agriculture System. International Journal of
Science, Strategic Management and Technology, 02(11), 1-9. https://doi.org/10.55041/ijsmt.v2i11.024
License: This article is published under the Creative Commons Attribution 4.0 International License (CC BY 4.0), permitting use,
distribution, and reproduction in any medium, provided the original author(s) and source are properly credited.
International Journal of Science, Strategic Management and Technology
Volume 02 Issue 11 November-2026 | ISSN: 3108-1762 (Online) | Impact Factor: 3.8
An International, Peer-Reviewed, Open Access Scholarly Journal Indexed in recognized academic databases
© Author(s). This work is peer-reviewed, openly published, and permanently archived
This article is openly accessible and reusable with proper attribution.
https://ijsmt.org/ , Email: editor@ijsmt.org 2
III. PROPOSED SYSTEM
The application contains five main modules. The
disease-detection module analyses an uploaded plant
image using a MobileNetV2 -basedCNN. The crop -
prediction module uses a Random Forest classifier
.Fertilizer suggestions are produced through
predefined nutrient -related rules, while the soil -
analysis module interprets nitrogen, phosphorus and
potassium values. A conversational assistant, powered
by Ollama, responds to general farming questions
Stream lit connects these functions through a common
user interface.
IV. METHODOLOGY
A. Plant Disease Detection
The user uploads a plant image in JPG or PNG format.
The image is resized to 224 × 224 pixels, converted to
RGB and normalized so that pixel values fall between
0 and 1. A MobileNetV2 -based CNN then extracts
image features and predicts a possible disease category
with a confidence score.
The project report identifies the PlantVillage dataset as
the source used for disease-detection work.
B. Crop Prediction
The crop-prediction module uses seven input features:
nitrogen, phosphorus, potassium, temperature,
humidity, pH and rainfall. A Random Forest classifier
combines the predictions of multiple decision trees and
returns the crop class receiving the most votes. There
port identifies crop_recommendation.csv as the dataset
used for this module.
C. Fertilizer and Soil Analysis
The fertilizer module compares nutrient values with
predefined conditions and provides a corresponding
recommendation. The soil-analysis module uses NPK
values to assign a fertility category. Since these
outputs depend on the rules and thresholds configured
in the application, they should be treated as
preliminary guidance rather than a substitute for
laboratory testing or advice from an agronomist.
D. Agricultural Assistant
The assistant uses Ollama to respond to general
questions about farming practices, crop selection and
disease prevention. Its responses are intended to
provide accessible information. They may not
account for a particular farm’s location, soil history,
crop variety or current field conditions.
E. Application Workflow
A user opens the Stream lit application and selects the
required function. For disease detection, the user
uploads an image. For crop prediction and soil-related
functions, the user enters the requested numerical
values. The relevant model or rule-based module
processes the input and displays its result. The project
also includes SQLite for local data storage where
database persistence is configured.
V. IMPLEMENTATION
The system is developed in Python, with Streamlit
used to build the interface. TensorFlow/Keras supports
the deep-learning component, Scikit -learn is used for
the Random Forest classifier, and image processing.
SQLite3 provides local database functionality, while
Ollama powers the conversational assistant.
The modules are kept separate so that each function
can be maintained or improved without redesigning
the entire application.
VI. TESTING AND PERFORMANCE
The project report describes tests covering plant -image
upload and disease prediction, crop prediction from
numerical inputs, fertilizer recommendations, soil-fertility
classification and chatbot responses. It reports disease -
detection accuracy of approximately 85 –90% and crop -
prediction accuracy of approximately 88 –92%. Image
processing is reported to take around 2 –3 seconds, while
crop prediction takes less than one second.
These figures should be interpreted with care. The report
does not provide a detailed train -test split, confusion
matrix or independent field -validation results for the
stated accuracy ranges. Actual performance may change
with image quality, dataset coverage, crop variety, input
accuracy and growing conditions. Further testing with
data collected from different farms and regions is therefore
necessary.
International Journal of Science, Strategic Management and Technology
Volume 02 Issue 11 November-2026 | ISSN: 3108-1762 (Online) | Impact Factor: 3.8
An International, Peer-Reviewed, Open Access Scholarly Journal Indexed in recognized academic databases
© Author(s). This work is peer-reviewed, openly published, and permanently archived
This article is openly accessible and reusable with proper attribution.
https://ijsmt.org/ , Email: editor@ijsmt.org 3
VII. RESULTS AND DISCUSSION
The implemented application brings the five
agricultural support functions into one interface. Users
can upload a plant image or enter soil and
environmental values, then view the prediction or
recommendation produced by the selected module. The
project report records successful operation of the listed
test cases and describes the interface as simple and
responsive.
The main contribution is the integration of several
functions in a single software platform without
requiring dedicated IoT hardware. At the same time,
the reliability of the recommendations depends on the
quality of the underlying models, datasets and rules.
The reported accuracy figures are preliminary project
results and should not be interpreted as guaranteed
performance in real field conditions.
VIII. CONCLUSION AND FUTURE WORK
This project presents an integrated smart -agriculture
application that combines plant disease detection, crop
prediction, fertilizer recommendation, soil -fertility
analysis and AI -based agricultural assistant. By
bringing these functions together in a Streamlit
interface, the system offers users a convenient starting
point for exploring plant and soil information.
By bringing these functions together in a Streamlit
interface, the system offers users a convenient starting
point for exploring plant and soil information. The
reported implementation demonstrates how deep
learning, machine learning and rule -based methods can
be used within one application.
Future improvements identified in the project include
connecting IoT sensors for real -time soil and
environmental readings, adding weather prediction,
expanding the datasets, improving model evaluation,
deploying the application securely on a cloud platform,
and developing a mobile version with regional -
language support. Testing with real farm data and
review by agricultural experts will be important before
the system is used to guide field-level decisions
X. REFERENCES
[1] S. P. Mohanty, D. P. Hughes, and M.
Salathé, “Using Deep Learning for Image-
Based PlantDisease Detection,” Frontiers in
Plant Science, vol. 7, Art.no. 1419, 2016. doi:
10.3389/fpls.2016.01419.
[2] M. Sandler, A. Howard, M. Zhu, A.
Zhmoginov, and L.-C. Chen,“MobileNetV2:
Inverted Residuals and Linear Bottlenecks,” in
Proc. IEEE/CVF Conference on Computer
Vision and Pattern Recognition, 2018,pp. 4510–
4520. doi: 10.1109/CVPR.2018.00474.
[3] L. Breiman, “Random Forests,”
Machine Learning, vol. 45, pp. 5–32,
2001.doi: 10.1023/A:1010933404324.
[4] “Crop Recommendation System using
Machine Learning,” International Journal of
Engineering Research & Technology
(IJERT). Available:
https://www.ijert.org/crop-recommendation-
system-using-machine-learning.
[5] TensorFlow, “MobileNet models.”
Available:
https://www.tensorflow.org/api_docs/python/t
f/keras/applications/MobileNetV2.
[6] Scikit-learn developers, “Random Forest
Classifier.” Available: https://scikit-
learn.org/stable/modules/generated/sklearn.en
semble.RandomForestClassifier.html.
[7] Stream lit, “Streamlit
documentation.” Available:
https://docs.streamlit.io/.
[8] Ollama, “Ollama documentation.”
Available: https://docs.ollama.com/.
[9] Python Software Foundation, “Python
3 Documentation.” Available:
https://docs.python.org/3/.
[10] Python Software Foundation, “sqlite3 —
DB-API 2.0 interface for SQLite databases.”
Available:
https://docs.python.org/3/library/sqlite3.html.
[11] PlantVillage, “PlantVillage Dataset.”
Available: https://github.com/spMohanty/PlantVillage-
Dataset.
论文原文与出版信息
International Journal of Science Strategic Management and Technology