You are on page 1of 107

EL

Machine Learning for Soil and Crop Management


Prof. Somsubhra Chakraborty
Agricultural and Food Engineering Department
Indian Institute of Technology Kharagpur

PT Week 1: GENERAL OVERVIEW OF ML AND


DL APPLICATIONS IN AGRICULTURE
N
EL
• Big data
• Machine Learning, AI, and DL
• Traditional uses of ML
• Supervised, Unsupervised, Semi-supervised, and Reinforcement learning

PT
• Applications of ML in agriculture
• Precision agriculture
• ML for crop management
• ML for soil management
N
• Artificial Intelligence

EL
• Machine Learning
• Regression
• Classification
• Supervised
• Unsupervised

PT
• Reinforcement learning
• Precision agriculture
• Crop management
• Yield prediction
• Crop quality
• Soil management
N
EL
BIG DATA

• Big data is a field that


analyze, systematically
extract information

PT
from, or otherwise deal
with data sets that are
too large or complex to
be dealt with by
traditional data-
processing application
software.
N
EL
BIG DATA

• Big Data: large volume of data,


which is produced by digital
technologies

PT
• Needs large storage
capabilities in addition to
editing, analyzing, and
interpreting
• Interpretation of big data:
considerable potential to add
value for society,
environment, and decision-
N
makers
EL
PT
N
(CC BY-SA 3.0)
CHALLENGES OF BIG DATA
• Big data encompass challenges on account of their so-called “5-V”

EL
requirements
 Volume
 Variety
 Velocity

PT
 Veracity
 Value
• The conventional data processing techniques are incapable of
meeting: emergence of ML !!
N
EL
ARTIFICIAL

PT
INTELLIGENCE
N
AI

EL
• The theory and development of
computer systems able to perform
tasks normally requiring human
intelligence, such as visual perception,
speech recognition, decision-making,
and translation between languages

PT • Google (Web search)


• Recommendation system (YouTube,
Amazon, Netflix)
• Human speech understanding (Siri or
Alexa)
N
• Self-driving car (Tesla)
AI

EL
• AI systems work by ingesting large
amounts of labeled training data,
analyzing the data for correlations and
patterns, and using these patterns to

PT
make predictions about future states

• In this way, a chatbot that is fed examples


of text chats can learn to produce lifelike
exchanges with people, or an image
recognition tool can learn to identify and
N
describe objects in images by reviewing
millions of examples
AI

EL
Three cognitive skills:

• Learning

PT • Reasoning

• Self-correction
N
AI: Advantages

EL
1. Beneficial for detail-oriented jobs

2. Reduced time for data-heavy jobs

PT
3. Offers consistent results

4. AI-powered virtual agents:


efficient and always available
N
AI: Disadvantages

EL
1. Costly

2. Requires deep technical

PT
knowledge

3. Limited experts to build AI tools

4. Lacking generalization from one


N
task to another
MACHINE
LEARNING

EL
Machine learning is a branch of artificial
intelligence (AI) and computer science which
focuses on the use of data and algorithms to
imitate the way that humans learn, gradually

PT
improving its accuracy
N
MACHINE LEARNING

EL
• The term machine learning was coined in
1959 by Arthur Samuel, an American IBMer
and pioneer in the field of computer gaming
and artificial intelligence.

PT
• Objectives:
• To classify data based on models which
have been developed
• To make predictions for future
outcomes based on these models
N
ML vs. AI

EL
• ML learns and predicts based on
passive observations, whereas AI
implies an agent interacting with the

PT
environment to learn and take actions
that maximize its chance of
successfully achieving its goals
Judea Pearl in The Book of Why
N
DEEP LEARNING

EL
• Part of a broader family of machine
learning methods based on artificial
neural networks with representation

PT
learning
• Learning can be supervised, semi-
supervised or unsupervised
N
N
PT
EL
TRADITIONAL USES OF ML

EL
• No human expertise (mars navigation)
• Humans can’t explain their expertise (speech recognition)
• Customized models (personalized medicine)
• Big data-based models (genomics)

PT
N
TRADITIONAL USES OF ML

EL
• Stock market forecasting
• Credit Card fraud detection
• Contamination detection
• Image recognition

PT
N
TRADITIONAL USES OF ML

EL
PT
N
TYPES OF LEARNING
1. Supervised/ inductive learning

EL
– Given: training data + desired outputs (labels)
- ML task of learning a function that maps an input to an output
based on example input-output pairs
2. Unsupervised learning
– Given: training data (without desired outputs/ with unlabeled

PT
outputs)
3. Semi-supervised learning
– Given: training data + a few desired outputs (small number of
labelled outputs)
4. Reinforcement learning
– Rewards from sequence of actions (rewarding desired behaviors
N
and/or punishing undesired ones)

Based on slide by Pedro Domingos


SUPERVISED LEARNING: REGRESSION

EL
• Given (x1, y1), (x2, y2), ..., (xn, yn)
• Learn a function f(x) to predict y given x-y are numerical= regression

PT
N
EL
PT
Source: “Supervised machine learning in a nutshell” by Epochfail, Licensed under CC BY-SA 4.0(CC BY-SA 4.0)

SUPERVISED
• Given (x1, y1), (x2, y2), ..., (xn, yn)
LEARNING: • Learn a function f(x) to predict y given x-y are numerical= regression
REGRESSION
N
SUPERVISED LEARNING: CLASSIFICATION

EL
• Given (x1, y1), (x2, y2), ..., (xn, yn)
• Learn a function f(x) to predict y given y is categorical= classification

PT
N
https://www.oreilly.com/
SUPERVISED LEARNING

EL
• X can be multi-dimensional
• Each dimension= one attribute

x1
PT
N
x2
SUPERVISED LEARNING: CLASSIFICATION PERFORMANACE METRICS

EL
PT
1. True Positive (TP): The plant has a disease (1) and the model classifies this case as diseased (1)
2. True Negative (TN): The plant does not have a disease (0) and the model classifies this case as healthy (0)
3. False Positive (FP): The plant does not have a disease (0), but the model classifies this as diseased (1)
4. False Negative (FN): The plant has a disease (1), but the model classifies this case as healthy plant (0)

Benos et al. (2021)


N
SUPERVISED LEARNING: CLASSIFICATION PERFORMANACE METRICS

EL
PT
N
OTHER COMMON METRICS

EL
PT
N
UNSUPERVISED LEARNING

EL
• Given x1, x2,………., xn (without labels)
• Output hidden structure behind the x’s
• – E.g., clustering

PT
N
UNSUPERVISED LEARNING

EL
• Given x1, x2,………., xn (without labels)
• Output hidden structure behind the x’s
• – E.g., clustering

PT Cluster 1
N
Cluster 2

Interpretation/ Processing
UNSUPERVISED LEARNING

EL
• Unsupervised learning is a ML technique in which models are not
supervised using training dataset. Instead, models itself find the
hidden patterns and insights from the given data. It can be
compared to learning which takes place in the human brain while

PT
learning new things.
N
UNSUPERVISED LEARNING

EL
• Unsupervised learning cannot be directly applied to a
regression or classification problem because unlike
supervised learning, we have the input data but no
corresponding output data.

PT
• The goal of unsupervised learning is to find the underlying
structure of dataset, group that data according to
similarities, and represent that dataset in a compressed
N
format
UNSUPERVISED LEARNING: WHY IMPORTANT?

EL
1. UL is helpful for finding useful insights from the input data.
2. UL is much similar as a human learns to think by their own
experiences, which makes it closer to the real AI.
3. UL works on unlabeled and uncategorized data which make
unsupervised learning more important.

PT
4. More close to the real-world problems: we do not always
have input data with the corresponding output so to solve
such cases, we need unsupervised learning
N
UNSUPERVISED LEARNING: 2 TYPES

EL
-Method of grouping the
objects into clusters based on
similarity.

Clustering -Finds the commonalities


between the data objects and
categorizes them as per the
presence and absence of those
commonalities.

UL

PT
-Used for finding the
relationships between
variables in the large
database
Association -Determines the set of items
that occurs together in the
dataset
- Market analysis
N
RENFORCEMENT LEARNING

EL
• Deals with how intelligent agents can
take actions in an environment in
order to maximize the notion of From opsrules.com

cumulative reward.
• Used by

PT
various software
machines to find the best possible
and

behavior or path it should take in a


N
specific situation.
RENFORCEMENT LEARNING: EXAMPLE

EL
• We have an agent and a reward, with many hurdles in
between. The agent is supposed to find the best possible
path to reach the reward. The following problem explains
the problem more easily.

PT
N
https://www.geeksforgeeks.org/what-is-reinforcement-learning/
RENFORCEMENT LEARNING: EXAMPLE

EL
PT
N
FRAMING OF RENFORCEMENT LEARNING

EL
An agent takes actions in an
environment, which is
interpreted into a reward and
a representation of the state,
which are fed back into the
agent.

PT
Credit: Megajuice
N
RENFORCEMENT LEARNING

EL
• Input: The input should be an initial state from which the
model will start
• Output: There are many possible output as there are
variety of solution to a particular problem

PT
• Training: The training is based upon the input. The model
will return a state and the user will decide to reward or
punish the model based on its output.
N
• The model keeps continues to learn.
• The best solution is decided based on the maximum
reward. https://www.geeksforgeeks.org/what-is-reinforcement-learning/
RL vs. SL

EL
PT
N
https://www.geeksforgeeks.org/what-is-reinforcement-learning/
RENFORCEMENT LEARNING: 2 TYPES

EL
When an event, occurs
due to a particular
behavior, increases the
strength and the
Positive frequency of the
behavior. In other words,
it has a positive effect on

PT
behavior.

RL
Strengthening of a
behavior because a
Negative negative condition is
stopped or avoided.
N
POSITIVE AND NEGATIVE RL
POSITIVE NEGATIVE

EL
Advantages Maximizes Performance Increases Behavior

Sustain Change for a long Provide defiance to


period of time minimum standard of
performance

Disadvantages Too much Reinforcement can


lead to overload of states
Only provides enough to
meet up the minimum

PT
which can diminish the behavior
N results

https://www.geeksforgeeks.org/what-is-reinforcement-learning/
RENFORCEMENT LEARNING APPLICATIONS

EL
• Robotics for industrial automation
• ML and data processing
• Training systems that provide custom instruction and
materials according to the requirement of students.

PT
N
Image Credit: ICAPlants
ML SYSTEM

EL
PT
N
Benos et al. (2021)
ML SYSTEM

For converting complex raw data

EL
into a suitable state, a pre-
processing effort is required.
This usually includes:
(a) data cleaning for removing
inconsistent or missing items

PT
Benos et al. (2021)
and noise
(b) data integration, when many
data sources exist and
(c) data transformation, such as
normalization and discretization
N
ML SYSTEM

The extraction/selection feature

EL
aims at creating or/and
identifying the most informative
subset of features in which,
subsequently, the learning model
is going to be implemented

PT
Benos et al. (2021)
throughout the training phase
N
ML SYSTEM

Feedback loop: Adjustments

EL
pertaining to the feature
extraction/selection unit as well
as the pre-processing one that
further improves the overall
learning model’s performance.

PT
Benos et al. (2021)

Testing/ validation: Unknown


samples are imported to the
trained model, which are usually
represented as feature vectors.
N
DEEP LEARNING

EL
• A subfield of ML
• Utilizes an alternative architecture via
shifting the process of converting raw
data to features (feature engineering)
to the corresponding learning system

PT
Benos et al. (2021)
• The feature extraction/selection unit is
absent, resulting in a fully trainable
system
• Starts from a raw input and ends with
the desired output
N
DEEP LEARNING

EL
• Learning feature labels in
scenes: Convolution networks

PT From Le Cun group,


N
Hinton group, Ng group
CHALLENGES OF MODERN AGRICULTURE

EL
• Increasing food demand
• Global population explosion
• Climate changes
• Natural resource depletion

PT
• Alteration of dietary choices
• Safety and health concerns
N
CONSEQUENCES

EL
• Pressure on Agricultural sector
• Need: optimization of agricultural practices
without putting extra environmental burden

PT
• PRECISION AGRICULTURE!
N
PRECISION AGRICULTURE

EL
‘Precision agriculture is a management strategy that gathers, processes
and analyzes temporal, spatial and individual data and combines it with
other information to support management decisions according to
estimated variability for improved resource use efficiency, productivity,
quality, profitability and sustainability of agricultural production’

PT
N International Society of Precision Agriculture
EL
PT
PRECISION AGRICULTURE
N
PRECISION AGRICULTURE (PA)
• Increased production
• Reduced labor time

EL
• Effective management of
GIS crop
yield maps
fertilizers and irrigation

Recommendation
(Fertilizer,

PA
Vegetation Irrigation,
Pesticides)

VRT

PT Algorithms
for biotic
and abiotic
stress
detection
N
Cycle of precision agriculture. GIS, Geographic information system; VRT, variable
rate technology.

Adopted from Abdullahi, H., Sheriff, R., 2017. Case study to investigate the adoption of precision agriculture in nigeria
using simple analysis to determine variability on a maize plantation. J. Agric. Econ. Rural Dev. 3 (3), 279–292 (Abdullahi
and Sheriff, 2017), p. 284. Under CCBY license—Creative Commons-licensed research.
BENEFITS OF AGRICULTURAL MODERNIZATION

EL
Sustainability

Maximum
productivity

PT Safe
environment
N
FOUR PILLARS OF SMART FARMING

EL
• Optimal natural resource management
• Conservation of the ecosystem
• Adequate services development

PT
• Use of modern technologies
N
INFORMATION AND COMMUNICATION TECHNOLOGY (ICT)
• Farm management information systems

EL
• Humidity and soil sensors
• Accelerometers
• Wireless sensor networks
• Cameras

PT
• Drones
• Low-cost satellites
• Online services
• Automated guided vehicles
N
APPLICATION OF ML IN AGRICULTURE

EL
PT
N
Benos et al. (2021)
APPLICATION OF ML IN AGRICULTURE

EL
Geographical distribution
of the contribution of
each country to the
research field focusing on
machine learning in
agriculture

PT
N
Benos et al. (2021)
APPLICATION OF ML IN AGRICULTURE

EL
PT
N
Benos et al. (2021)
ML MODELS GIVING THE BEST OUTPUT

EL
PT
N
Benos et al. (2021)
THE 10 MOST INVESTIGATED CROPS USING ML MODELS

EL
PT
N
Benos et al. (2021)
MOST USEFUL FEATURES

EL
PT
N
Benos et al. (2021)
CROP MANAGEMENT

EL
• Involves versatile aspects that originated from the combination of farming
techniques in the direction of managing the biological, chemical and
physical crop environment with the aim of reaching both quantitative and
qualitative targets.
• Using advanced approaches to manage crops, such as

PT
• Yield prediction
• Disease detection
• Weed detection
• Crop recognition
• Crop quality detection
N
YIELD PREDICTION

EL
• One of the most challenging topic in Agriculture
• An accurate model = help the farm owners to take informed
management decisions on what to grow towards matching the
crop to the existing market’s demands

PT
• Very difficult task!
• Determined by several factors such as environment, management
practices, crop genotypic and phenotypic characteristics, and
their interactions
• Require comprehensive data + powerful ML models
N
YIELD PREDICTION
A novel deep neural

EL
network (DNN)-based
methodology is proposed
to predict the future crop
yield based on historical
yields and greenhouse
environmental

PT
parameters (e.g., CO2
concentration,
temperature, humidity,
radiation, etc.)
information
N
Gong et al. (2021), Sensors
YIELD PREDICTION
The proposed method is

EL
based on the hierarchical
integration of the
recurrent neural
network (RNN) and
temporal convolutional
network (TCN), which are

PT
both the current
state-of-the-art DNN
architectures for
temporal sequence
processing
N
Gong et al. (2021), Sensors
YIELD PREDICTION
4 main parts

EL
1. Normalization part
2. Recurrent neural network
part
3. Temporal convolutional
network part
4. Fully connected layer part

PT
N
Gong et al. (2021), Sensors
YIELD PREDICTION

EL
PT
Accumulated tomato fruit yield (g/m2)
N
Gong et al. (2021), Sensors
YIELD PREDICTION

EL
Gong et al. (2021), Sensors
Ground truth tomato
fruit yield values and
predicted ones for
testing datasets

PT
N
YIELD PREDICTION
A deep CNN-LSTM

EL
model for both
end-of-season and
in-season soybean
yield prediction

PT
N
Sun et al. (2019), Sensors
YIELD PREDICTION
Scatter plots of

EL
end-of-season
predicted vs.
observed yield
from 2011 to 2015

PT
N
Sun et al. (2019), Sensors
DISEASE DETECTION

EL
• Crop disease= a major threat in Agril production system
• Deteriorates yield quality, quantity at production, storage and
transportation level
• Pose significant risk to food security

PT
• Timely identification of disease pest is important
• Historically, disease detection was conducted by expert
agronomists by performing field scouting= time-consuming and
solely based on visual inspection.
N
DISEASE DETECTION

EL
• An optimal algorithm for detecting damaged and diseased potato tubers moving
along a conveyor belt using computer vision systems

PT
N
Korchagin et al. (2021), Agronomy
DISEASE DETECTION

EL
• An optimal algorithm for detecting damaged and diseased potato tubers moving
along a conveyor belt using computer vision systems

PT
N
Korchagin et al. (2021), Agronomy

Identification of potato tubers by the Viola-Jones method


DISEASE DETECTION

EL
• Stochastic gradient
descent momentum
optimizer and a
discount momentum

PT
(DM) deep learning
optimizer for plant
diseases identification
N
Sun et al. (2021), Applied sciences
DISEASE DETECTION

EL
• Tuning hyperparameters

PT
N
Sun et al. (2021), Applied sciences
DISEASE DETECTION

EL
• ResNet50: ResNet50 has
50 layer deep CNNs with
skip connections for
image classification.

PT
• DenseNet121:
DenseNet121 is a 121-
layer deep CNNs with
dense connections
N
Sun et al. (2021), Applied sciences
WEED
DETECTION

EL
• As a result of their prolific
seed production and

PT
longevity, weeds usually grow
and spread invasively over
large parts of the field very
fast, competing with crops for
the resources, including
space, sunlight, nutrients, and
water availability
N
WEED
DETECTION

EL
• Weed losses are expected to
reach 11 billion USD per year

PT
in India, ranging from 13.8%
in transplanted rice to 76% in
soybean; in which, weeds
contribute the highest
potential loss, accounting for
34% of all biotic stressors,
followed by insects of 18%
N
and diseases of 16%

Mohidem et al. (2021), Agriculture


WEED
DETECTION

EL
• Besides, weeds frequently
arise sooner than crops

PT
without having to face natural
enemies, a fact that adversely
affects crop growth
• Weed control is necessary
N
WEED
DETECTION

EL
• Mechanical control:
ineffective

PT
• Chemical control: herbicides
• Herbicides: costly and
environmental hazardous
• Long term herbicides: crop
resistance
N
WEED
DETECTION

EL
• Smart Farming:
• Remote sensing

PT
• UGV
• UAV
• UAV based data collection and
classification: difficult
• ML, Imaging technologies, and non-
imaging spectroscopy: real-time
differentiation and localization of
target weeds, enabling precise
application of herbicides to specific
N
zones, instead of spraying the
entire fields
UAV SYSTEMS FOR WEED DETECTION

EL
PT
N
Mohidem et al. (2021), Agriculture
UAV SYSTEMS FOR DISEASE DETECTION

EL
PT
N
Pan et al. (2021), Sensors
MOBILE ROBOT PLATFORM FOR WEED DETECTION

EL
PT Cutulle and Maja (2021), Italian journal of Agronomy
N
CROP RECOGNITION

EL
• Automatic recognition of crops has gained considerable attention
in several scientific fields, such as plant taxonomy, botanical
gardens, and new species discovery
• Plant species can be recognized and classified via analysis of

PT
various organs, including leaves, stems, fruits, flowers, roots, and
seeds
N
Benos et al. (2021)
CROP RECOGNITION

EL
• leaf-based plant recognition seems to be the most common
approach by examining specific leaf’s characteristics like color,
shape, and texture
• With the broader use of satellites and aerial vehicles as means of

PT
sensing crop properties, crop classification through remote
sensing has become particularly popular
• The advancement on computer software and image processing
devices combined with ML has led to the automatic recognition
and classification of crops
N
Benos et al. (2021)
CROP QUALITY

EL
• Important for market
• High quality produce= higher earning to the farmer
• Developing decision support systems can aid farmers in taking
appropriate management decisions for increased quality of

PT
production
• ML applications: selective harvesting is a management practice
that may considerably increase quality
N
Benos et al. (2021)
TRADITIONAL SOIL TESTING

EL
PT
N
TRADITIONAL SOIL TESTING

EL
PT
N
N
PT
EL
PROXIMAL SENSOR+ML FOR SOIL CLASSIFICATION

EL
PT
N
DIFFUSE REFLECTANCE SPECTROSCOPY
• Fast

EL
• Cost-effective (almost zero recurring cost)
• Needs minimumor no sample pretreatments
• High Throughput and Non-invasive
• Portable

PT • Parameters:
• OC, AvailableN, P, K
• pHandclay
• Moistureandheavymetals
N
Credit: D.C. Weindorf
DIFFUSE REFLECTANCE SPECTROSCOPY

EL
PT
N
Vestergaard et al. (2021), Sensors
PREDICTING SURFACE SOIL MOISTURE BY DRS

EL
PT
N
Zhu et al. (2010)
IMAGE PROCESSING + DL

EL
PT
N
IMAGE PROCESSING + DL

EL
PT
N
Swetha et al. (2020), Geoderma
IMAGE PROCESSING + DL

EL
PT
N
Swetha et al. (2020)
IMAGE PROCESSING + DL

EL
PT
N
Gorthi et al. (2021)
DIGITAL SOIL MAPPING

EL
PT
N
Gorthi et al. (2021)
MAPPING SOIL N AT SMALLHOLDER MAIZE FARM

EL
PT
N
Mashaba-Munghemezulu et al. (2021)
MAPPING SOIL N AT SMALLHOLDER MAIZE FARM

EL
PT
N
Mashaba-Munghemezulu et al. (2021)
• ML in agriculture: an ever-increasing topic worldwide

EL
• Requires for better crop and soil management

• Development of new ML tools: enabling smart agriculture

PT
• Sensors + ML: boon for resource-poor countries
N
• Benos, L., Tagarakis, A. C., Dolias, G., Berruto, R., Kateris, D., & Bochtis, D.
(2021). Machine Learning in Agriculture: A Comprehensive Updated Review.

EL
Sensors, 21(11), 3758.
• Gong, L., Yu, M., Jiang, S., Cutsuridis, V., & Pearson, S. (2021). Deep Learning
Based Prediction on Greenhouse Crop Yield Combined TCN and RNN. Sensors,
21(13), 4537.
• Gorthi, S., Swetha, R. K., Chakraborty, S., Li, B., Weindorf, D. C., Dutta, S., ... &
Majumdar, K. (2021). Soil organic matter prediction using smartphone-captured
digital images: Use of reflectance image and image perturbation. Biosystems

PT
Engineering, 209, 154-169.
• Swetha, R. K., Bende, P., Singh, K., Gorthi, S., Biswas, A., Li, B., ... & Chakraborty,
S. (2020). Predicting soil texture from smartphone-captured digital images and
an application. Geoderma, 376, 114562.
• Vestergaard, R. J., Vasava, H. B. B., Aspinall, D., Chen, S., Gillespie, A.,
Adamchuk, V., & Biswas, A. (2021). Evaluation of Optimized Preprocessing and
Modeling Algorithms for Prediction of Soil Properties Using VIS-NIR
N
Spectroscopy. Sensors, 21(20), 6745.
N
PT
EL

You might also like