Obviousness

To assess the obviousness of US patent 11030491 under 35 U.S.C. § 103, we must determine whether the differences between the claimed invention and the prior art would have been obvious to a person having ordinary skill in the art (POSITA) at the time of the invention (priority date: September 23, 2016). A POSITA in this field would likely be a data scientist or software engineer with expertise in machine learning (particularly deep learning and computer vision), Geographic Information Systems (GIS), and a foundational understanding of property assessment, risk analysis, or insurance underwriting. The motivation to combine prior art elements stems from the recognized need for more efficient and accurate automated property characteristic and condition assessment, as highlighted in the patent's background.

The core of the independent claims (Claim 1, 11, and 15) involves:

1. Obtaining imagery (aerial, potentially terrestrial, and shape maps).
2. Using machine learning to identify and classify property characteristics (e.g., roof shape).
3. Using machine learning to classify the condition of those characteristics (e.g., roof condition).
4. Utilizing these classifications to determine a risk estimate or replacement cost.

Combination of Prior Art for Obviousness

A primary combination that would render the claims of US11030491 obvious would involve:

• Krizhevsky et al. ("ImageNet Classification with Deep Convolutional Neural Networks", 2012): Discloses the use of Convolutional Neural Networks (CNNs) for robust image classification and feature extraction.
• Lin et al. ("Network In Network", 2014): Describes the Network in Network (NIN) model, an advanced deep learning architecture for image classification, noted for superior performance and efficiency compared to conventional CNNs.
• General availability of imagery and mapping data: This includes widely accessible aerial imagery (e.g., Google® Earth, United States Geological Survey, Geospatial Information Authority (GSI) of Japan) and shape map data (e.g., GSI, Zenrin Co. Ltd. of Japan), as noted in the patent's description.
• Established image processing techniques: General knowledge of image analysis methods such as color histogram analysis and pattern recognition for assessing image qualities or identifying defects.
• Existing need in insurance/property assessment industry: The patent itself identifies the problem of manually assessing property characteristics for risk exposure databases and the desire to "automate predictive analytics" and "more accurately estimate" risk of damage due to disaster.

Motivation to Combine

A POSITA would be strongly motivated to combine these elements for the following reasons:

1. Efficiency and Automation: The manual assessment of property characteristics and conditions is time-consuming and prone to human error. The recognized "promise of deep learning" is "replacing human identification of features with efficient algorithms for unsupervised or semi-supervised feature learning and hierarchical feature extraction." Therefore, a POSITA would naturally seek to apply advanced machine learning to automate this process.
2. Improved Accuracy: Deep learning algorithms like CNNs (Krizhevsky et al.) and NIN (Lin et al.) had already "demonstrated superior performance outcome to conventional CNN processing" in image classification tasks by the priority date. A POSITA would be motivated to leverage these cutting-edge techniques to enhance the accuracy of identifying property characteristics (e.g., roof shape) from visual imagery.
3. Comprehensive Assessment: Moving beyond just identifying a property characteristic (e.g., "this is a roof"), to assessing its condition (e.g., "this roof is in poor condition") is a logical and obvious extension for anyone tasked with risk or value assessment. Existing image processing techniques, such as color histogram analysis (as explicitly mentioned in the patent for condition analysis) or pattern recognition (e.g., identifying missing shingles), are well-known tools for detecting variations or defects in images. Applying these to regions identified by deep learning as specific property features would provide more granular and valuable data for risk estimation.
4. Integration of Data Sources: The widespread availability of aerial imagery, terrestrial imagery, and shape maps from various public and private sources (Google Earth, GSI, QGIS, Zenrin, etc.) would motivate a POSITA to integrate these diverse data streams. Techniques like overlaying shape maps with aerial images to "confirm location," "aid cropping," or "correct or compensate for alignment errors or inconsistencies" are fundamental practices in GIS and remote sensing for ensuring data accuracy before analysis. This would be a routine step to improve the reliability of the machine learning inputs.
5. Direct Business Application: Once property characteristics and their conditions are automatically classified, applying this information to calculate a "risk estimate of damage" or a "replacement cost" is a direct and obvious application to solve a known problem in the insurance or real estate industries. The patent itself outlines the use cases for estimating damage risk, repair costs, or confirming repairs. This represents a straightforward application of derived data to existing actuarial or appraisal models.

Obviousness of Specific Claim Elements

• Obtaining images (aerial, shape maps, terrestrial): These sources were publicly available and routinely used for geographic and property-related information prior to the patent's priority date.
• Identifying and classifying property characteristics using deep learning (CNN or NIN): Krizhevsky et al. and Lin et al. teach these specific deep learning models for image classification. A POSITA would choose these, or similar, for feature identification in imagery. The patent explicitly mentions both as potential classifiers.
• Classifying condition using machine learning (e.g., color histogram analysis): While Krizhevsky or Lin don't specifically teach "condition," the concept of using machine learning or image processing to assess qualities like wear, damage, or degradation from images is well-established. Color histogram analysis, as taught in the patent, is a basic image processing technique for analyzing pixel intensity distributions, which can be correlated to condition (e.g., uniformity, discoloration, or patches can indicate poor condition).
• Determining risk estimate or replacement cost from classifications: This is a direct application of the classified data (property type + condition) to solve a known problem in the insurance and real estate industries. The motivation is clear: more accurate and efficient risk assessment and cost estimation.
• Image preparation steps (overlaying shape maps, assessing orthogonality, cropping/resizing): These are standard practices in GIS and remote sensing to ensure the accuracy and usability of imagery for analysis. A POSITA would routinely perform such steps to optimize the input for machine learning models.
• Determining preferred image type based on characteristic: Knowing that certain features are better seen in aerial vs. terrestrial images (e.g., "housing siding features may be more easily detected in terrestrial imagery") is a common-sense design choice for anyone working with various image types.
• Accessing known property characteristics (e.g., age, elevation): Combining image-derived data with existing structured data about a property is a standard data fusion technique to enhance the accuracy of analytical models.

In conclusion, the claimed invention in US11030491 represents an obvious combination of known machine learning techniques (deep learning for image classification), widely available imagery sources, and standard image processing methods, applied to the known problem of inefficient and less accurate property assessment for risk and cost estimation in the insurance and real estate sectors. A POSITA, motivated by the desire for automation, increased accuracy, and comprehensive data integration, would have been led to combine these elements to achieve the claimed results.