How To: Classify a Feature with Multiple Labels Using ArcGIS Pro

This video tutorial demonstrates how to use a multi-label feature classification deep learning model with ArcGIS Pro. The multi-label model is capable of classifying a single feature with multiple labels, unlike traditional classifiers that assign only one label per feature. The video provides a practical example of using this model to survey residential properties for the presence of swimming pools and solar panels, a task that would be time-consuming without automation. The workflow involves exporting a label dataset, training a model with that data, and then using the trained model to classify objects. The video also covers the process of exporting high-resolution tiles, adjusting training parameters, and running the inference to classify features. The result is a more efficient method for identifying and categorizing features with multiple attributes.

Takeaways

• 📚 Multi-label feature classifiers can assign multiple labels to a single feature, unlike traditional classifiers which assign a single label.
• 🏠 An example application is classifying residential properties for features like pools and solar panels, which can save time and resources.
• 📈 The workflow involves exporting a label dataset, training a model, and then using that model for inference.
• 📊 The label dataset should have a specific format, with fields named in a way that reflects the labels (e.g., 'multi-label_pool' for pools).
• 🖼️ High-resolution imagery is used to create tiles for training, ensuring each feature is distinct and isolated.
• 🤖 The deep learning model is trained using the exported data, with options to adjust parameters like batch size and learning rate.
• 💻 GPU acceleration is recommended for training the model to speed up the process.
• 🔍 The 'Classify Objects Using Deep Learning' tool is used for inference, classifying features based on the trained model.
• ⚙️ Parameters such as batch size and environment settings should be consistent between training and inference for optimal results.
• 📉 The model can accurately classify features as having no labels (e.g., neither a pool nor solar panels) by using a 'none' label.
• 📈 The output shows confidence levels for each label, indicating the model's certainty in its classification.
• 🎓 The video provides a comprehensive guide on using multi-label feature classifiers in ArcGIS Pro for more efficient feature classification.

Q & A

What is a multi-label feature classifier in the context of ArcGIS Pro?

A multi-label feature classifier is a deep learning model that can assign multiple labels to a single feature, unlike traditional classifiers which assign only one label per feature.

Why would a city authority use a multi-label feature classifier?

City authorities might use a multi-label feature classifier to survey residential properties and determine which properties have features like swimming pools and solar panels, which can help in assessing property rates, understanding the affinity level, and other related tasks.

How does the multi-label classifier differ from a single-label classifier in terms of feature classification?

The multi-label classifier can classify a single feature with multiple labels, such as both a swimming pool and solar panels, whereas a single-label classifier can only assign one label to a feature.

What are the three parts of the workflow for using a multi-label classifier in ArcGIS Pro?

The workflow consists of exporting the label dataset, training a model using the exported dataset, and performing inference using the trained model.

How does the attribute table in ArcGIS Pro represent multi-label features?

The attribute table uses a specific format for multi-label features, with field names starting with 'multi-label' followed by an underscore and the label name, which helps the tool to identify and use the label names.

What is the purpose of the 'multi-label none' field in the attribute table?

The 'multi-label none' field is used to represent land parcels that do not have any of the labeled features, such as neither a swimming pool nor a solar panel, ensuring the model learns to recognize the absence of features as well.

What are the parameters that can be adjusted when exporting a label dataset for deep learning in ArcGIS Pro?

Parameters include the input raster, input feature layer, type of data to create, metadata format (multi-label tiles), buffer around the feature, cell size for high resolution, and the name for the output.

How can the number of epochs and batch size be adjusted during the training of a deep learning model in ArcGIS Pro?

These parameters can be adjusted in the 'Train Deep Learning Model' tool, where you can increase the number of epochs and batch size depending on the available memory or GPU memory.

What is the role of GPU in the training and inference of a deep learning model in ArcGIS Pro?

A GPU (Graphics Processing Unit) can be selected as the processing type during both training and inference to accelerate the process by leveraging its parallel processing capabilities.

How does the 'Classify Objects Using Deep Learning' tool in ArcGIS Pro utilize the trained model for inference?

The tool uses the trained model to classify objects by providing the input features and the model definition, allowing it to predict the presence of multiple labels for each feature.

What are the potential outcomes of using a multi-label classifier on a land parcel in ArcGIS Pro?

The outcomes can include correct classification of land parcels with multiple features (e.g., pool, solar panel), single features (e.g., only pool or only solar panel), or no features (none), each with associated confidence scores.

Outlines

00:00

📚 Introduction to Multi-Label Feature Classification in ArcGIS Pro

The video begins with an introduction to a multi-label feature classification deep learning model within ArcGIS Pro. It explains the concept of multi-label classification, which allows features to have multiple labels simultaneously, unlike traditional single-label classifiers. The presenter uses the example of city authorities surveying residential properties to identify those with swimming pools and solar panels, a task that can be significantly expedited using this model. The workflow is outlined in three parts: exporting the label dataset, training the model using the dataset, and performing inference with the trained model. The video also discusses the importance of the specific format for field names in multi-label classification and demonstrates how to export a label dataset using the 'Export Training Data for Deep Learning' tool, including setting parameters like cell size and metadata format.

05:01

🤖 Training and Inference with a Multi-Label Deep Learning Model

The second paragraph delves into the training process of the deep learning model. It details how to use the 'Train Deep Learning Model' tool, providing the input training data from the previously exported label dataset. The presenter discusses customizable parameters such as batch size, chip size, learning rate, and the choice of different model backbones. Additionally, it touches on the option to split the validation set and the importance of selecting a GPU for the environment if available. The video then moves on to the inference stage, showcasing the 'Classify Objects Using Deep Learning' tool. It demonstrates how to apply the trained model to new data, adjust the batch size, and run the inference to classify features. The presenter concludes by showing the results of the inference, highlighting the model's ability to accurately identify features with multiple labels, such as properties with both swimming pools and solar panels, as well as those with none of these features.

Keywords

Multi-label feature classifier

A multi-label feature classifier is a type of machine learning model that can assign more than one class or label to a single feature. Unlike traditional classifiers that assign a single label, this model recognizes the possibility of multiple attributes being present in one instance. In the context of the video, it is used to classify land parcels as having features like swimming pools and solar panels simultaneously.

ArcGIS Pro

ArcGIS Pro is a powerful geographic information system (GIS) application developed by Esri that allows users to create, analyze, and share geographic information. It is used in the video to demonstrate how to apply a deep learning model for classifying features with multiple labels, showcasing its capabilities in handling spatial data and machine learning integration.

Deep learning model

A deep learning model refers to artificial neural networks with multiple layers designed to learn and make decisions based on data inputs. These models are capable of complex pattern recognition and are widely used in image and speech recognition, among other applications. In the video, a deep learning model is trained to classify land parcels based on the presence of features like swimming pools and solar panels.

Land parcel polygons

Land parcel polygons are geographical data representations used to denote specific areas of land, often for the purpose of mapping, property assessment, or planning. In the video, these polygons are the primary features being classified by the multi-label feature classifier to determine the presence of amenities such as swimming pools and solar panels.

Attribute table

An attribute table is a data structure in GIS that holds the non-spatial data (attributes) related to the spatial features in a GIS project. It is similar to a spreadsheet or database table and is used to store information about each feature, such as its identification, type, and characteristics. In the video, the attribute table is used to show how land parcels are marked with '1' for the presence or '0' for the absence of features like swimming pools and solar panels.

Export training data

Exporting training data is the process of preparing and saving a dataset that will be used to train a machine learning model. This data typically includes input features and corresponding labels. In the context of the video, the process involves setting parameters such as tile size and metadata format to create a dataset suitable for training the multi-label deep learning model.

Model training

Model training is the phase in machine learning where an algorithm is taught to make predictions or decisions based on a training dataset. The training process involves adjusting the model's parameters to minimize error. In the video, the deep learning model is trained using the exported land parcel data to learn the features associated with the presence of swimming pools and solar panels.

Inference

Inference in the context of machine learning refers to the process of applying a trained model to new, unseen data to make predictions or classifications. It is the final step after training where the model's performance is evaluated. In the video, inference is used to classify new land parcels using the trained deep learning model to determine if they have swimming pools, solar panels, or neither.

Batch size

Batch size in machine learning is the number of training examples used in one iteration. It is a critical parameter that can affect the model's training speed and performance. Larger batch sizes can lead to faster training but may require more memory. In the video, the presenter mentions adjusting the batch size based on the available GPU memory during model training.

GPU

A GPU, or Graphics Processing Unit, is a specialized electronic circuit designed to rapidly manipulate and alter memory to accelerate the creation of images in a frame buffer intended for output to a display device. In the context of the video, a GPU is used to provide the necessary computational power for training the deep learning model due to its ability to handle parallel processing tasks efficiently.

Confidence level

The confidence level in machine learning is a measure of how certain the model is about its predictions. It is often expressed as a percentage or a probability value. A higher confidence level indicates a stronger belief in the model's prediction. In the video, the model's confidence level is discussed when it classifies land parcels, with high confidence levels indicating accurate classifications.

Highlights

The video demonstrates how to use a multi-label feature classifier deep learning model with ArcGIS Pro.

Multi-label models can classify features with multiple labels, unlike single-label classifiers.

An example is given where a feature can be classified as having both a swimming pool and solar panels.

The process is applicable for city authorities to survey residential properties for amenities like swimming pools and solar panels.

The workflow consists of three parts: exporting the label dataset, training a model, and performing inference.

Features are marked with '1' for the presence and '0' for the absence of a label.

The attribute table uses a specific format for multi-label fields, identified by 'multi-label M_L_' prefix.

Exporting the label dataset involves using the 'Export Training Data for Deep Learning' tool with specific parameters.

High-resolution imagery is used to ensure each tile contains a single land parcel for training.

The deep learning model is trained using the exported data with customizable parameters.

The number of epochs, batch size, and learning rate can be adjusted during model training.

Different model backbones can be experimented with for improved performance.

A GPU can be specified in the environment settings for faster training.

The 'Classify Objects Using Deep Learning' tool is used for inference with the trained model.

Batch size and environment settings should match those used for the export dataset.

Inference results show the presence of features like swimming pools and solar panels with confidence scores.

The model can correctly classify land parcels without any features as 'none'.

The video concludes with a demonstration of the multi-label feature classifier's practical application in ArcGIS Pro.