Obviousness

Obviousness Analysis of US Patent 11195058 under 35 U.S.C. § 103

This analysis assesses the obviousness of US patent 11195058 (hereinafter '058 patent) under 35 U.S.C. § 103, considering the state of the art as of the priority date of September 23, 2016.

Person Having Ordinary Skill in the Art (POSA)

A person having ordinary skill in the art (POSA) in the field of the '058 patent would likely possess a bachelor's or master's degree in computer science, electrical engineering, or a related field, with several years of experience in machine learning, image processing, remote sensing, and/or geographic information systems (GIS), particularly in applications related to property assessment or risk analysis. This POSA would be familiar with various machine learning models, including neural networks and deep learning architectures, as well as standard image acquisition and processing techniques.

Summary of Core Claims

The '058 patent generally claims a system and method for automatically categorizing property characteristics and their repair/maintenance conditions through aerial imagery analysis. Key aspects include:

• Obtaining aerial imagery of a property.
• Identifying features corresponding to a property characteristic.
• Analyzing features (e.g., using deep learning like NIN) to determine a property characteristic classification (e.g., roof shape).
• Analyzing a region of the image (e.g., using machine learning like color histogram analysis) to determine a condition classification (e.g., good/bad).
• Determining a risk estimate of damage or replacement cost based on both classifications.
• Pre-processing steps such as obtaining and overlaying shape maps, assessing orthogonality, and selecting optimal imagery.
• Integration with other known property data and real-time processing.

Identified Prior Art References (from '058 Patent Text)

The '058 patent's own background and detailed description acknowledge several components of the prior art relevant to the invention:

• [A] Risk Exposure Databases and Visual Imagery for Property Characteristics: The patent explicitly states, "A risk exposure database contains a compilation of as many building properties or characteristics relevant to insurance as possible... Some of these characteristics can only be assessed by on-site inspections or by official documentation, but others can be measured using visual imagery." This establishes the known problem of gathering property data for insurance/risk assessment and that visual imagery was a recognized source. It also highlights the importance of "roof condition, roof shape, roof covering, roof anchors, roof equipment, cladding, and pounding" for these databases.
• [B] Deep Learning for Image Classification: The '058 patent refers to well-established deep learning methodologies for image classification, specifically citing "Alexnet" (Krizhevksy et al., 2012) and "Network in Network (NIN)" (M. Lin et al., 2014) as machine-learning models. It notes that "By using deep learning algorithms and sample datasets, computers can learn to distinguish and classify a wide range of characteristics to high levels of accuracy, often surpassing the recognition levels of human beings." It further states, "One of the promises of deep learning is replacing human identification of features with efficient algorithms for unsupervised or semi-supervised feature learning and hierarchical feature extraction."
• [C] Relationship between Property Characteristics/Condition and Disaster Vulnerability: The patent acknowledges that "Each roof shape has a unique response and damage vulnerability to different natural perils like earthquake or wind." This demonstrates the known motivation to classify characteristics and conditions for risk estimation.
• [D] Geographic Information Systems (GIS) and Sources of Aerial/Shape Map Imagery: The patent describes obtaining aerial imagery from remote databases like "Google® Earth images by Google, Inc." or "NTT Geospace Corporation of Japan," and publicly owned organizations such as "the Geospatial Information Authority (GSI) of Japan" or "the United States Geological Survey." It also mentions "Open Source Geographic Information System (GIS) such as QGIS by the Open Source Geospatial Foundation (OSGeo)" for collecting imagery. For shape maps, it mentions "Geospatial Information Authority of Japan or Zenrin Co. Ltd. of Japan."

Obviousness Combinations and Motivation to Combine

Combination 1: Core Automation for Property Characteristic and Condition Assessment for Risk Estimation (A + B + C + D)

A POSA, as of September 23, 2016, would have been motivated to combine the known elements of prior art [A], [B], [C], and [D] to achieve the automated property characteristic and condition assessment described in the '058 patent.

• [A] Knowledge of Risk Exposure Databases and the Need for Property Data: A POSA would recognize the existing need for efficient and accurate methods to populate risk exposure databases with property characteristics and conditions, traditionally a labor-intensive process involving manual assessment or on-site inspections.
• [B] Deep Learning for Image Classification: With the advancements in deep learning demonstrated by Alexnet (2012) and NIN (2014), a POSA would readily appreciate that these powerful image classification techniques could be applied to automate the extraction of characteristics from visual imagery. The patent itself states the inventors "recognized that deep learning methodology could be applied to risk exposure database population to analyze aerial imagery and automatically extract characteristics of individual properties, providing fast and efficient automated classification of building styles and repair conditions."
• [C] Understanding of Property Feature Vulnerability: The known correlation between property characteristics (like roof shape and condition) and their susceptibility to damage from natural disasters would provide a strong motivation for a POSA to accurately identify and classify these features and their conditions to improve risk estimates.
• [D] Availability of Aerial/Shape Map Imagery and GIS Tools: The widespread availability of aerial imagery from various commercial and public sources, combined with GIS tools for geographic data handling, would make these images a natural and accessible data source for automated analysis.

Motivation to Combine: The motivation for a POSA to combine these references is clear: to automate and improve the efficiency and accuracy of gathering property characteristic and condition data for insurance and risk assessment. By leveraging the proven capabilities of deep learning (B) to analyze readily available aerial imagery (D), the tedious and error-prone manual identification of property features (A) and their conditions could be replaced with an automated system. This automation would directly feed into the known process of estimating damage risk based on these characteristics (C), leading to more accurate and timely risk assessments. The combination represents an obvious application of known technological advances (deep learning) to solve a known problem (efficient property data collection for risk assessment) using readily available data sources (aerial imagery).

Combination 2: Image Pre-processing for Enhanced Machine Learning Analysis (A + B + D + General Image Processing Knowledge)

The '058 patent's pre-processing steps, such as obtaining shape maps, overlaying them with aerial images, determining boundary matches, and assessing orthogonality, would also be obvious to a POSA when applying machine learning to aerial imagery for property analysis.

• [A], [B], [D]: As described in Combination 1, these references establish the context of using machine learning on aerial imagery for property analysis.
• General Image Processing Knowledge: A POSA would have general knowledge of standard image processing techniques used to prepare images for analysis, especially for machine learning applications where input quality can significantly impact results. This includes techniques for geometric correction, alignment, and cropping.

Motivation to Combine:

• Overlaying Shape Maps for Alignment and Cropping: When applying machine learning to analyze features within specific property boundaries, a POSA would be motivated to accurately define the area of interest. Overlaying a known shape map (D) with an aerial image (D) is a standard GIS practice to "confirm location of a particular property" and "to match properties with images." The patent itself states this overlay "can be used in aiding in cropping the aerial image 102 c to focus analysis on a particular property location 102 b." This is a logical step to improve the precision of the input data for the machine learning model (B) and ensure the analysis is focused on the correct property features for the risk exposure database (A).
• Assessing and Correcting Orthogonality: Aerial images often contain perspective distortions. For accurate measurement and feature extraction, especially when comparing against map data or performing detailed condition analysis, correcting these distortions to generate a "true orthophoto" is a well-known process in remote sensing. A POSA seeking to use aerial imagery for precise property characteristic and condition analysis would understand the necessity of geometrically corrected images to ensure accuracy and consistency of the data used for machine learning (B). The patent explicitly mentions that "an aerial image representing a normal orthophoto angle may not be directly centered upon the planning map block" and that "the aerial image can be geometrically corrected to obtain a true orthophoto version of the aerial image." This indicates it was a known problem and solution.
• Selecting Best Quality/Preferred Image Type: When multiple sources of imagery are available (D), it is an obvious engineering choice to select the "best quality image for use in condition analysis," balancing factors like "clarity, completeness, and recency." Similarly, understanding that "Different property characteristics may be discerned based upon whether the aerial image is captured in two-dimensional or three-dimensional format" and that "Housing siding, for example, is more easily detected in terrestrial imagery 102 d and/or three-dimensional aerial imagery 120 c than in two-dimensional aerial imagery 102 c" would motivate a POSA to select the appropriate image type for a given characteristic to optimize the machine learning analysis.

Combination 3: Specific Condition Analysis Techniques (B + General Knowledge of Image Processing Techniques like Color Histogram Analysis)

• [B] Deep Learning for Characteristic Identification: A POSA would use deep learning (B) to identify and classify property characteristics (e.g., roof type).
• General Knowledge of Image Processing Techniques (e.g., Color Histogram Analysis, Pattern Recognition Algorithms): Beyond deep learning, a POSA would be familiar with other standard image processing techniques suitable for analyzing specific attributes like condition. The patent explicitly states that for condition analysis, "the machine learning analysis in some examples, can include two-dimensional color histogram analysis or three-dimensional color histogram analysis. In other embodiments, the machine learning analysis may be performed using pattern recognition algorithms (e.g., determining missing fence posts or missing/misaligned rooftop shingles)." The patent also illustrates how "good condition histograms" differ from "poor condition histograms" for rooftops.

Motivation to Combine: Once a property characteristic (e.g., a roof) has been identified and classified by type (e.g., hipped roof) using deep learning (B), a POSA would be motivated to employ appropriate and efficient image analysis techniques to assess its condition. Color histogram analysis is a well-known technique for analyzing pixel intensity distributions and color uniformity, which are direct indicators of degradation or damage in visual imagery. Similarly, pattern recognition algorithms are standard for detecting specific defects like missing shingles or fence posts. These methods are efficient for assessing condition features after the primary characteristic has been identified. Therefore, applying these known image processing techniques to assess the condition of identified features, potentially in conjunction with or as a distinct step after deep learning for characteristic identification, would be obvious to a POSA.

Conclusion

Considering the knowledge explicitly stated within the '058 patent regarding the state of the art in risk exposure databases, deep learning, image acquisition, and property assessment needs, the claimed methods and systems would have been obvious to a person having ordinary skill in the art as of the priority date of September 23, 2016. The patent describes an obvious application of known machine learning and image processing techniques to a well-understood problem, with clear motivations for each combination of elements to automate and improve existing processes.