No Name Date of Publishing

MIS-ME: A Multi-modal Framework for

1 2024
Soil Moisture Estimation

Crop Selection and Yield Prediction using

2 2023
Machine Learning Approach

Deep Learning Precision Farming:

3 Grapes and Mango Leaf Disease 2021
Detection by Transfer Learning

Deep Neural Networks to Detect Weeds

4 from Crops in Agricultural Environments 2021
in Real-Time

Farm-Scale Crop Yield Prediction from

5 2021
Multi-Temporal Data
 Generative Adversarial Networks for
6 2022
Image Augmentation in Agriculture

Potato Leaf Disease Detection Using

7 2023
Machine Learning

Recommending and Predicting Crop

8 Yield using Smart Machine Learning 2023
Algorithm

Tomato Crop Disease Classification

9 Using Pre-Trained Deep Learning 2018
Algorithm

Machine-Learning Based Retrieval of Soil

10 Moisture at High Spatio-Temporal Scales 2020
Using CYGNSS and SMAP Observations

Image Analysis and Detection of Tea

11 2021
Leaf Disease using Deep Learning
 A Hybrid Computer Vision enabled GAN
12 2022
Model for Pest and Disease Detection

A Review: Classification and Detection

Of Plants Diseases Using Machine
13 2022
Learning And Soft Computing
Techniques

An Efficient Method for Leaf Diseases

14 Detection Using Deep Learning 2023
Technique

Real-Time Plant Disease Detection Using

15 2023
Mobile Phone Images

A Benchmark of Deep Learning Models

16 2021
for Multi-leaf Diseases for Edge Devices

A CNN and Random Forest Fusion with

Optimized Convolutional Layers for
17 2024
Accurate Disease Identification in Rice
Using Machine Learning
 A Comparative Analysis of Potato Leaf
Disease Classification with Big Transfer
18 2022
(BiT) and Vision Transformer (ViT)
Models

A Comparative Study of Hyperparameter

19 Tuning in Deep Learning Models using 2024
Bayesian Optimization and XAI

Revolutionizing Farming: GAN-Enhanced

20 Imaging, CNN Disease Detection, and 2024
LLM Farmer Assistant
Summary Methodology Used
CNN models (ResNet18,
MobileNetV2, InceptionV3,
Multi-modal framework, MIS-ME, for estimating soil
EfficientNetV2)
moisture using both meteorological data and real-
world images captured from ground stations. The
Multi-modal learning using CNN-
framework aims to improve soil moisture prediction,
based image feature extraction
which is essential for precision agriculture. It
and meteorological data
demonstrates that combining image-based and
processing, with three fusion
statistical weather data leads to better estimation
techniques (concatenation,
accuracy compared to using either method alone.
hybrid loss, and learnable
parameters).

Machine learning-based system for predicting the best aïve Bayes for crop classification
crop selection and estimating crop yield based on and Random Forest Regression
environmental and soil parameters. It uses for yield prediction, with dataset
classification models for crop prediction and regression preprocessing, feature selection,
models for yield prediction. and model evaluation.

he paper proposes a deep learning-based system for Transfer learning with AlexNet
detecting and classifying diseases in grape and mango CNN, image preprocessing,
leaves using transfer learning. It employs AlexNet to segmentation, and classification
automatically extract features and classify diseased using deep learning on a dataset
leaves with high accuracy. of leaf images.

This paper reviews the use of deep learning models for

Deep learning-based image
real-time weed detection in agricultural fields. It
recognition with CNN, YOLO,
discusses the effectiveness of neural networks in
Faster R-CNN, and Mask-RCNN
identifying and classifying weeds while highlighting
for weed detection and
challenges like occlusion, lighting variations, and
classification.
dense vegetation.

The paper presents a deep hybrid neural network Hybrid deep learning model
model that predicts farm-scale crop yield using multi- combining Convolutional Neural
temporal data, including Sentinel-2 satellite images, Networks (CNNs) and Recurrent
weather data, and farm management records. Neural Networks (RNNs)
Generative Adversarial Networks (GANs) for image
augmentation in agriculture, focusing on enhancing GAN architectures such as
machine learning models for tasks like plant health DCGAN, CycleGAN, StyleGAN,
detection, weed recognition, and postharvest quality and WGAN f
assessment.

This paper proposes a CNN-based approach for

Pre-trained VGG16 CNN model
detecting potato leaf diseases, specifically early blight
with transfer learning, data
and late blight. The model is fine-tuned on a dataset of
preprocessing, augmentation,
potato leaf images to classify them as healthy or
and fine-tuning.
diseased.

Machine learning models

including Support Vector
This paper proposes a Smart Machine Learning Machine (SVM), Artificial Neural
Algorithm (SMLA) for recommending suitable crops Networks (ANN), Random Forest
and predicting yield based on soil and weather (RF), Multivariate Linear
parameters. The model helps farmers make informed Regression (MLR), and K-Nearest
decisions to improve productivity. Neighbors (KNN) for crop
recommendation and yield
prediction.

This paper explores the use of pre-trained deep Transfer learning using AlexNet
learning models, AlexNet and VGG16, for classifying and VGG16 CNN architectures,
tomato leaf diseases. It analyzes the impact of with image preprocessing,
hyperparameters like batch size, weight, and bias augmentation, and fine-tuning
learning rates on classification accuracy. for disease classification.

This paper proposes a machine learning approach to Random Forest machine learning
estimate surface soil moisture using data from the model using CYGNSS reflectivity,
CYGNSS satellite constellation. It enhances soil NDVI, vegetation water content,
moisture retrieval accuracy by integrating remote elevation, and soil texture data
sensing parameters and reference datasets. for soil moisture retrieval.

CNN (LeNet); image resizing to

Focuses on detecting tea leaf diseases using LeNet 256x256; data augmentation
CNN to classify diseases from images. Aims to help (affine transformations,
farmers identify diseases early and avoid losses. rotations); image annotation with
border boxes.
 GAN-based models for data
augmentation; image
Proposes a GAN-based model for pest and disease
preprocessing (resizing,
detection, improving accuracy by generating synthetic
renaming); feature extraction
data for limited datasets.
(edge detection, texture
analysis).

CNN-based classification; various

Reviews machine learning and soft computing
ML and soft computing
techniques for plant disease classification, highlighting
techniques for automatic feature
CNN's effectiveness for feature extraction.
extraction and classification.

CNN classifier with k-means

Proposes an efficient deep learning method combining clustering and HSI color
CNN with k-means clustering and HSI color conversion conversion; contrast
for leaf disease detection. enhancement for preprocessing;
deep learning for classification.

GAN-based data augmentation;

CNN for classification;
Focuses on real-time detection of plant diseases using preprocessing (resizing,
mobile phone images and ML techniques, without IoT renaming, contrast
or satellite imagery. enhancement); feature
extraction (texture, color
transformations).

Evaluates MobileNets, MnasNet,

Benchmarks deep learning models for multi-leaf EfficientNets Lite, InceptionNets,
disease detection suitable for edge devices with and ResNets using TensorFlow
limited resources. Lite on Raspberry Pi 3; post-
training quantization applied.

CNN architecture with three

convolutional layers integrated
Introduces a unified method for rice disease detection
with Random Forest; stratified
by integrating CNNs and Random Forest.
sampling and data augmentation
(rotation, flips) applied.
 Trained BiT and ViT models on
potato leaf dataset;
Compares Big Transfer (BiT) and Vision Transformer
hyperparameter tuning via grid
(ViT) models for potato leaf disease classification.
search; developed with
TensorFlow.

Uses Bayesian Optimization on

Presents a comparison study on hyperparameter
CNN, ResNet, MobileNet,
tuning of deep learning architectures for early rice
InceptionNet, and RegNet;
plant disease detection using Bayesian Optimization
employs LIME for explainability
and XAI.
and interpretable results.

Dataset of 2075 original images

Integrates Generative AI (GAN), CNNs, and Large expanded to 3075 using GAN;
Language Models (LLMs) for crop disease recognition CNN for disease detection;
and farmer assistance. LangChain and LLAMA for AI-
powered farmer chatbot.
Research Gaps Results

Existing models rely either on Mean Absolute Percentage Error

meteorological data or high-cost satellite (MAPE) of 10.14%,
imagery, but none effectively combine outperforming traditional
real-world soil images with weather data unimodal approaches by 3-5%
for soil moisture estimation. in accuracy.

Naïve Bayes classifier achieved

Traditional methods rely on historical 99.39% accuracy for crop
averages and do not incorporate real-time prediction, while Random Forest
weather data or advanced machine Regression achieved a Mean
learning techniques to improve accuracy Absolute Error (MAE) of 0.64
in crop yield prediction. and an R² score of 0.96 for yield
prediction.

xisting models lack real-time

99.03% accuracy for grape leaf
implementation for disease detection in
disease detection and 89.86%
diverse lighting and field conditions,
for mango leaf disease
limiting their usability for farmers in real-
detection.
world scenarios.

Existing methods struggle with real-world accuracy for weed detection

variations in lighting, occlusion, and dense was 99.29% using transfer
vegetation, making them less reliable for learning on pre-trained CNN
real-time applications. models

Previous studies focused on regional-scale

predictions, but few addressed farm-scale
Mean Absolute Error (MAE) of
yield variations due to management
76 kg per 1000 m²,
decisions and localized environmental
conditions.
Agricultural image datasets are often
small and imbalanced, limiting the
effectiveness of deep learning models; 97% accuracy in plant disease
GANs can help but require improvements detection.
in generating realistic and diverse images
under varying environmental conditions.

Existing models lack real-time deployment

97.4% accuracy, with high
and optimization for diverse
precision (1.00), recall (0.44),
environmental conditions affecting potato
and an F1-score of 0.61
leaf disease detection.

Existing models lack a comprehensive

approach that integrates soil, weather, 95% accuracy in crop
and fertilizer data while optimizing recommendation and yield
recommendations for better yield prediction,
prediction accuracy.

Existing models do not optimize

hyperparameters effectively for tomato AlexNet achieved 97.49%
disease classification, and further accuracy, while VGG16
improvements in execution time and obtained 97.23%
generalization are needed.

The model achieved an

Existing satellite-based soil moisture unbiased root-mean-square-
retrieval methods lack high spatial and difference (ubRMSD) of 0.0523
temporal resolution, limiting their m³/m³ for soil moisture
effectiveness in real-time agricultural estimation and 0.042 m³/m³
applications. when compared to SMAP data,
demonstrating high accuracy.

Blight leaf disease achieved

Limited to specific tea leaf diseases;
highest accuracy; model
generalization to other crops like tomato
outperforms traditional
and mulberry; potential methodological
backpropagation networks
enhancements.
based on ROC evaluation.
 Improved detection accuracy
Need for diverse real-world datasets;
through GAN-based data
exploration of advanced GAN architectures
augmentation; F-score
for better augmentation; evaluation on
identified as a reliable metric
mobile-acquired images.
considering precision and recall.

CNN-based methods show high

Lack of real-time application studies;
classification accuracy;
limited exploration of hybrid ML models;
suggests potential for real-time
performance testing across diverse
disease detection with further
conditions is needed.
optimizations.

High classification accuracy

Limited real-world scenario testing; need
achieved with CNN and k-
for advanced preprocessing optimization;
means clustering; contrast
potential to apply methodology to broader
enhancement improved model
plant species.
performance.

Expected improvements in
Limited dataset challenges; adapting detection accuracy and
drone-based methodologies to mobile robustness when tested with
images; optimizing preprocessing for real- mobile images; evaluation
world conditions. using accuracy, precision,
recall, and F-score.

MobileNet V3 achieved 96.58%

Further optimization of models for diverse accuracy, with inference and
edge devices and generalization across initialization times of 127 ms
various crops. and 11 ms respectively, and 7.4
MB memory usage.

Achieved 83.25% micro-

Testing model on diverse rice varieties and
average accuracy; Brown Spot
real-field conditions; further balancing of
detection at 93.19% accuracy;
complexity and interpretability.
False Smut recall at 89.80%.
 BiT-M-R50x3 model achieved
Exploring performance on larger, more 100% training precision, recall,
diverse datasets and real-world F1-score; test scores: 97.329%
environments. precision, 97.624% recall,
97.469% F1-score.

Bayesian optimization
Assessing generalization of
enhanced deep learning
hyperparameter tuning across different
models’ efficiency; LIME model
datasets and crops; integration of
provided insights into model
additional XAI techniques.
behavior.

CNN model reached 0.997

Assessing scalability of the integrated accuracy and 0.0167 loss; AI
system; testing GAN-generated images in chatbot integration aimed to
real-world disease identification scenarios. enhance farmer decision-
making and accessibility.
Dataset

Cropland images and

meteorological data

India Agriculture Crop

Production, District Wise
Rainfall Normal, and Crop
Recommendation datasets.

PlantVillage dataset and

locally acquired grape and
mango leaf images.

DeepWeeds, CropDeep,

Sentinel-2 satellite imagery,

weather data, farm reports,
New Plant Diseases Datasethttps://www.kaggle.com/datasets/vipoooool/new-plant-diseases-dataset

PlantVillage dataset with

13,262 segmented images
of tomato leaves, including
six disease classes and a
healthy class.

CYGNSS satellite data, Soil

Moisture Active Passive
(SMAP) data, and
International Soil Moisture
Network (ISMN) ground-
based measurements.

80 images from the plant

village dataset (blister
blight, red scab, red leaf
spot, leaf blight).
Synthetic images generated
via GANs; real-world drone
imagery (proposed for
future mobile-based
adaptation).

Various publicly available

plant disease image
datasets; focus on
classification accuracy and
feature extraction.

Images of leaves processed

for contrast, resized, and
clustered; dataset specifics
not mentioned explicitly.

Mobile phone-captured
plant leaf images under
diverse real-world
conditions; GAN-augmented
synthetic images for
robustness.

Large-scale dataset from

PlantVillage, evaluated on
Raspberry Pi 3.

Dataset includes Bacterial

Blight, Sheath Blight, False
Smut, Stem Rot, and Brown
Spot.
Dataset of healthy and
diseased potato plant
leaves, including Early
Blight, Healthy, and Late
Blight.

Rice plant disease datasets

evaluated across multiple
deep learning models.

Original dataset of 2075

images expanded to 3075
using GAN techniques.
No Dataset Used for Dataset URL
1 Crop health https://www.kaggle.com/datasets/misrakahmed/vegetable-imag
2 Image data set soil https://huggingface.co/datasets/KaraAgroAI/Drone-based-Agricu
3 Crop and weed dataset https://github.com/cropandweed/cropandweed-dataset
4 Soil Moisture dataset https://github.com/osu-complex-systems/mis-me
5 Dataset for Crop Pest and Disease DetectionCrop Pest and Disease Detection
6 Plant Disease Expert Dataset https://www.kaggle.com/datasets/sadmansakibmahi/plant-disea
 ts/misrakahmed/vegetable-image-dataset
/KaraAgroAI/Drone-based-Agricultural-Dataset-for-Crop-Yield-Estimation
d/cropandweed-dataset
x-systems/mis-me

ts/sadmansakibmahi/plant-disease-expert
Can a smartphone image reveal the hidden details of soil
hydration and overall crop health? Is it possible to
empower farmers with immediate, data-driven insights
through advanced image analysis? We propose a novel
framework that leverages U-Net and DeepLab v3+ for
precise semantic segmentation, complemented by
YOLOv5 for robust object detection, to transform farm
images into detailed agronomic data. By bridging the gap
between established research and practical application,
our model delivers critical measurements and actionable
recommendations from a single image. Field evaluations
demonstrate that our approach achieves high accuracy
and real-time performance across diverse agricultural
conditions. This system optimizes resource management
and enhances sustainable practices, improving crop
guiding sustainable practices and enhancing crop
productivity.
productivity.