FUTURE of AGRICULTURE

FUTURE of AGRICULTURE

Quick Summary

• Data Related Usecases - #Forecast yield, water/nitrogen needs
• #Vision Related - Predict disease, patterns/weather info
• Mix of Skills - Agriscience, Data, AI, Transfer Learning, Customizing to new markets / similar
• Recommendations - Derive insights/spot risks plus options to mitigate / prescriptive options
• Model climate adaptation changing weather to recommend suitable crops
• Impact - 13 countries, 500 crops, 10k varieties
• Interesting - Model to predict a new variety of crop yield
• Intelligent cloud for sustainable agriculture.
• Using Images, Data, Text, historical data everything to build intelligence

Detailed Insights

• Massive dataset / Best position to build knowledge graph
• Apps for Farmers
• Scaling digital solution to 500 crops, 10k varieties
• AI models for 22 commodities, 13 countries
• Provide Data and Infra
• Breadth and Depth in Data from data collected
• Challenging applying model in unseen regions
• General AI capability and Knowledge Graph
• Wheat in India vs Nigeria vs Canada
• Iterating on Capability
• Agriscience, Capability Science
• 13 countries, 2.2 billion hectare scale
• Country scale error predictions Nigeria (Wheat prediction)
• Model predicted disease options
• Modell hypertuned for different farms / conditions / transfer learning
• predict yeild model in crops
• water / nitrogen update
• predict disease / optimize water
• Model an asset
• Sustainability score
• Predict disease - model climate side of risk
• Spread and breadth of dataset
• Domain knowledge Vision, Agriscience
• Models to detect Cloud Detection vs Cloud Shadow
• Model climate adaptation, risks, recommendations
• Predict with minimum dataset
• Model to predict new variety of crop yeild
• Partnering with cloud providers
• Cloud for intelligent agriculture
• Partner with industry and solve problems
• Apps / Platform

Ref - Link

Knowledge graphs can incorporate both structured (for example, coming from a spreadsheet, or precision agriculture equipment) and unstructured data (a twitter feed, images, YouTube video, bulletin board information, books etc.) Knowledge graphs can be successful and valuable if they can uncover new insights by automatically incorporating new data sources, understanding the context, finding new connections, and continuously evolving and learning.

Keep Exploring!!!

Research Paper Reads - Fruit Freshness

Paper #1 - Deep Learning Based Classification System For Recognizing Local Spinach

• Five types of spinach. (a)Jute Spinach, (b)Malabar Spinach, (c)Red Spinach, (d)Taro Spinach, (e)Water Spinach.
• Because our dataset contains different sizes of the images. 224x224x3 is the input shape of our model

Paper #2 - Automatic Plant Cover Estimation with Convolutional Neural Networks

• This dataset contains 7,200 images with 750 training and 150 validation images per class and is therefore also balanced
• Our network consists of two main components: a feature extractor backbone and a network head
•  The backbone consists of one of the abovementioned standard classification networks potentially in conjunction with a Feature Pyramid Network to increase the output resolution

Paper #3 - Fruit Freshness Grading Using Deep Learning

• Fruit colour is derived from natural pigments when ripening, enzymatic and non-enzymatic browning reactions lead to the formation of water-soluble dark colours
• Visual characteristics, e.g., shape, wholeness, spots, bruises, and blemishes, can reflect the speed of fruit deterioration
• The consistency of physical shape may indicate the thickness of fruits that may have implications of its capability to defend against diseases
• Geometric changes are a frequently observed result of fruit degradation
• Texture is another important measurement of the level how a fruit has decayed
• Fruit texture, colour and shape are three important visual features for fruit quality grading

• Added random noises follow the sequential order: Random brightness adjustment, random contrast, and random erasion

• In total, there are (approximate) 4,000 images collected with each type of fruit about 700
• The freshness grading is scaled from 0.0 to 10.0 with 0.0 indicating total corruption and 10.0 for total freshness 
• We define the fruits being harvested as absolute freshness with a numerical level description of 10.0

Paper #4 - Machine Vision based Fruit Classification and Grading - A Review

• Some extraction methods like Speeded Up Robust Features (SURF), Histogram of Oriented Gradient (HOG) and Local Binary Pattern (LBP) 
• Features of fruits like color, size, shape and texture
• Automatic sorting system that can perform fast, save time and reduce manual labor
• The basic steps of the automatic image-based fruit grading are: fruit image recognition, fruit object recognition

Color features extraction methods broadly fall in two categories:

1. Global methods (global color histogram, histogram intersection, image bitmap)

2. Local methods (local color histogram, color correlogram, color difference histogram)

# Fitting K-NN to the Training set

from sklearn.neighbors import KNeighborsClassifier

classifier = KNeighborsClassifier(n_neighbors = 5, metric = 'minkowski', p = 2)

classifier.fit(X_train, y_train)

# Predicting the Test set results

y_pred = classifier.predict(X_test)

Paper #5 - Deep Learning for Automatic Quality Grading of Mangoes: Methods and Insights

• Augment the CNN classifier with a convolutional autoencoder
• The rationale of considering such networks is as follows: 1) the presence of autoencoder
• forces the network to remember essential information for reconstruction when extracting features for classification, thereby having a regularizing effect; and, 2) the latent features learned by the network could benefit other downstream tasks as they contain the compressed information for reconstruction;

Data Augmentation

• Horizontal or vertical flip, each with 50% probability;
• Brightness, –20 to +20%;
• Contrast, –10 to +10%;
• Rotation, –20 to 20 degrees;
• Zoom in/out, 0.8x to 1.25x

The ConvAE-Clfs consist of 3 components:

• A convolution-based encoder that compresses an image into a latent vector;
• A convolution-based decoder that reconstructs the image from the latent vector and some intermediate features;
• A fully-connected classifier that takes the latent vector as input and gives the class prediction.

The proposed convolutional autoencoder-classifiers were shown to have no clear advantage over the single-task CNNs, but the result should be verified with larger datasets and more related tasks

Paper #6 - Deep Learning and Machine Vision for Food Processing: A Survey

• An MVS includes two main parts to enable objective and non-destructive food evaluation: 1) acquiring and 2) processing

• Fruit and Vegetable Identification Using Machine Learning for Retail Applications
• GuavaNet: A deep neuralNet: A deep neural network ar al network architecture for automatic sensory evaluation  to predict the degree of acceptability for Guava by a consumer 

Keep Exploring!!!

Leaf Detection - Top K Techniques - Reads

Paper #1 PL@NTNET APP IN THE ERA OF DEEP LEARNING

Key Notes

• Observations are stored within a NoSQL document storage called CouchDb
• CNN architecture is the inception model
• The number of species (i.e. classes) in January 2017 was about 10K and the number of training images about 332K
• Species filtering is applied based on the checklist of species
• activated within the app (e.g. West Europe, North Africa, South America, etc.)
• By default, this checklist is automatically chosen according to the geo-location of the mobile device, but the user also has the possibility to select another one manually
• Similarity Search - This content-based image retrieval is performed through a hashing-based approximate nearest neighbors search algorithm applied on top of the 1024-dimensional feature vectors extracted by the last hidden layer of the fine-tuned CNN

Paper #2 - Fine-grained recognition of plants from images

Key Notes

• A number of approaches is based on the popular local binary patterns (LBP)
• Fast Features Invariant to Rotation and Scale of Texture (Ffrst)
• "One versus All" classifcation scheme is used for multi-class classifcation

• The Foliage leaf dataset by Kadir
• The Swedish leaf dataset
• The Leafsnap dataset

Paper #3 - Neural Network Application on Foliage Plant Identification

Key Notes

• Polar Fourier Transform that proposed by Zhang [18] has properties that are very useful for represents shape of objects, including leaf of plants
• There are three kinds of geometric features involved as shapes features: slimness ratio, roundness ratio, and dispersion. 
• Color features on a leaf can be extracted by using statistical calculations such as mean, standard deviation, skewness, and kurtosis

Leaf Type Classification

Paper #4 - An Online Algorithm for Large Scale Image Similarity Learning

Key Notes

• Here we focus on a weaker supervision signal: the relative similarity of different pairs
• we extract similarity information from pairs of images that share a common label
• we extract similarity information from pairs of images that share a common label or are retrieved in response to a common text query in an image search engine

Paper #5 - Large Scale Local Online Similarity/Distance Learning Framework based on Passive/Aggressive

Paper #6 - Comparison of Image Matching Techniques 

• Blob detection technique
• Template matching
• SURF feature extraction 

Paper #7 - Large Scale Online Learning of Image Similarity Through Ranking

Key Notes

• OASIS is both fast and accurate at a wide range of scales
• Similarity information is extracted from pairs of images that share a common label or are retrieved in response to a common text query
• OASIS can be trained on more than two million images within three days on a single CPU
• OASIS learned similarity show that 35% of the ten nearest neighbors of a given image are semantically relevant to that image
• Algorithm that uses triplets of images
• Color histograms are obtained by K-means clustering
• Local Binary Pattern
• LMNN - Large Margin Nearest Neighbor Classification

Paper #8 - Sparse online learning of image similarity

tf2_semantic_approximate_nearest_neighbors

More Reads

• Simple and fast method to compare images for similarity
• Sampling Wisely: Deep Image Embedding by Top-k Precision Optimization
• image-match
• Image Deduplicator (imagededup)
• Large Scale Local Online Similarity/Distance Learning Framework based on Passive/Aggressive
• Image similarity model
• Hashing Learning
• Fast Near-Duplicate Image Search using Locality Sensitive Hashing
• Querying Similar Images with TensorFlow
• Image classification with Keras and deep learning
• K-Nearest Neighbors Hashing
• A Survey on Deep Hashing Methods
• Image similarity estimation using a Siamese Network with a triplet loss
• Finding similar images using Deep learning and Locality Sensitive Hashing
• Fingerprinting Images for Near-Duplicate Detection
• LSH for near-duplicate image detection

Keep Exploring!!!

Leaf / Plant Detection - Reads :)

Never stop collecting / identifying/analyzing perspectives :).

Paper #1 - Real-world plant species identification based on deep convolutional neural networks and visual attention

Key Notes

• Data augmentation method for deep learning
• Crop the image in terms with visual attention
• Considering flowers and fruits of plants are seasonal, some researchers believe that leaves are more suitable for identification
• Samples is also strict

• PlantCLEF. There are different view types and the samples are close to realistic scenarios.

• Image segmentation is carried out for generating the regions of interest (ROI) for recognition
• Comparisons between original images and final attention cropping results.

Paper #2 - Deep Learning in Agriculture: A Survey

Key Notes

• High occlusion, depth variation, and uncontrolled illumination, including high color similarity between fruit/foliage
• Rotations, cropping, scaling, transposing, mirroring
• DetectNet CNN
• Faster Region-based CNN, DetectNet CNN

Paper #3 - PlantDoc: A Dataset for Visual Plant Disease Detection

Key Notes

• The PlantVillage dataset contains images taken under controlled settings. 
• Final dataset having a total of 27 classes spanning over 13 species with 2,598 images
• model which can detect a leaf in an image and then classify it into the particular classes

Paper #4 - Leaf Classification Using Shape, Color, and Texture Features

Key Notes

• Texture, on its own does not have the capability of finding similar images, but it can be used to classify textured images from non-textured ones 
• Texture features can be extracted by using various methods. Gray-level occurrence matrices (GLCMs), Gabor Filter, and Local binary pattern (LBP) 

• Fourier descriptors, slimness ratio, roundness ratio, and dispersion are used to represent shape features

Paper #5 - Multi-Organ Plant Classification Based on Convolutional and Recurrent Neural Networks

Key Notes

• Feature engineering approaches such as Scale-invariant feature transform (SIFT), Bag of Word (Bow), Speeded-Up Robust Features (SURF), Gabor, Local Binary Pattern (LBP)

Take multiple images and deduce plant, flower, leaf

Paper #6 - Two-View Fine-grained Classification of Plant Species

• Three levels of abstraction: family, genus, and species
• The input of the SCNN model is the whole leaf image characterizing a global
• view in terms of problem representation. The output of this stage is a ranked list of the top-K genus candidates
• SCNN takes into account global features extracted from the entire leaf image (shape and color), while in the second view, local features based on texture and the plant veins are considered. 
• The output of the first stage is a ranked list of the top-k genus candidates.
• In the second stage, given the top-k genus candidates found in the first stage, a fine classification considering only the plant species which belong to such a genus is performed

• A coarse-to-fine classification is performed considering the hierarchical botanic taxonomy
• e. Finally, the genus (coarse classification) and species (fine classification) are combined to produce a final ranked list of the k-best hypotheses of plant species.

LifeCLEF 2015 and LeafSnap datasets.

Paper #7 - Fine-Grained Visual Classification of Plant Species In The Wild: Object Detection as A Reinforced Means of Attention

Key Notes

• We introduce a bottom-up approach based on detecting plant organs and fusing the predictions of a variable number of organ-based species classifiers
• Object detection has mostly had its use in plant analysis constrained to invasive species detection
• First, an object detector identifies and localizes plant organs, including leaves, flowers, fruit, stems, and regions with a high volume of leaves, termed “highdensity leaves” (HDL)
• These regions of interest (ROIs) are then individually passed into an organ-based species classifier
• Leaf, flower, fruit, stem, HDL

Paper #8 - COMPUTER-AIDED INTERPRETABLE FEATURES FOR LEAF IMAGE CLASSIFICATION

Key Notes

• i) Convert original image to RGB (Red-Green-Blue) image, ii) Gray scaling, iii) Gaussian smoothing, iv) Binary thresholding, v) Remove the stalk, vi) Closing holes, and vii) Resize the image.

• Gray-scaling is the process of converting an image to shades of gray from other colour spaces like RGB
• Image smoothing techniques help in smoothing images and removing intensity edges.
• Thresholding is a segmentation technique that is used to separate the foreground from its background

Paper #9 - A Review of Visual Descriptors and Classification Techniques Used in Leaf Species Identification

• Leaf Snap Dataset
• The Middle European Wood Database
• The Smithsonian Leaf Dataset
• The Austrian Federal Forest (AFF) Dataset
• Pl@ntNet

More Reads

• Flower Categorization using Deep Convolutional Neural Networks
• A Leaf Recognition Algorithm for Plant Classification Using Probabilistic Neural Network
• Large-Scale Plant Classification with Deep Neural Networks

Keep Exploring!!!

Day #205 - Short ML Talks - Agriculture

Talk #1 - WACV18: DeepWheat: Estimating Phenotypic Traits from Crop Images with Deep Learning

Key Lessons

• Working on Aerial Image

• Automate Inspection for Bio-Mass
• Plant counting high rise picture
• Biomass Estimations

• Non-Linear Regression
• CNN for both
• Plant Counting, Leaf Counting 

Talk #2 - Implementation of Deep Learning in Agriculture Crop Identification

Key Lessons

• Identification of crop from satellite Images with crop based Deep Learning Models
• Google map satellite imagery
• Extract Image tiles and label them
• Vegetation Index with Chlorophyl Content
• Multispectral Imagery for Crop Health Analysis

Talk #3 - WACV18: Recognition of Pollen-bearing Bees from Video using Convolutional Neural Network

Key Lessons

• Ecological aspects Bees Disappearing
• Dataset setup
• Identification and Activity Analysis

Next Talks

WACV18: An Animal Detection Pipeline for Identification
Face Liveness Detection Based on Perceptual Image Quality Assessment Features 
WACV18: Crowd Counting With Minimal Data Using Generative Adversarial Networks
WACV18: Workshop: CDBR: Cross-Domain Biometric Recognition Overview
WACV18: Workshop: CV-AAL: Calorific Expenditure Estimation Using Deep Convolutional Network Features
WACV18: Workshop: CV-AAL: Assessing Pain Levels From Videos Using Temporal Convolutional Networks

Happy Mastering DL!!!

Day #197 - Tech Talks and ML based Ideas - Poverty Detection / Slum Detection using Deep Learning

Talk #1 - Neal Jean, " "Combining satellite imagery and machine learning to predict poverty"

Key Summary

• Global Poverty 1.9$ per day
• Need to identify poor people, Surveys are expensive
• Satellite images for classifying locations
• Infer Social-Economic Indicators

• Transfer Learning used for this task

• Learn from where you have more training data
• Nighttime light intensity for economic development

• Gaussian Process on top of CNN

Talk #2 - Using Convolutional Networks and Satellite Imagery to Identify Patterns

Key Summary

• Compare regions within the city
• Intelligent Energy and Infrastructure benchmarking
• Compare various areas of City
• Land Use Classification
• DeepSat Dataset
• Supervised Learning Approach
• pysatml, pysatapi
• VGG16 for Classification
• Pretrain on DeepSat Data and finetune them

Related Work
Slum Segmentation and Change Detection :
Mumbai-slum-segmentation

Next Task
awesome-human-pose-estimation
Human Pose Estimation with TensorFlow
keras-openpose-reproduce
Realtime Multi-Person Pose Estimation
Snagging Parking Spaces with Mask R-CNN and Python

Happy Mastering DL!!!