7  Classification2

7.1 Summary

7.1.1 Object based image analysis (OBIA)

Considering shapes based on the similarity(homogeneity) or difference(heterogeneity) of the cells(superpixels)

• Simple Linear Iterative Clustering(SLIC): the most common method for superpixel generation

• R packages: Supercells, SegOptim

7.1.2 Sub pixel analysis

• Also termed: Sub pixel classification, Spectral Mixture Analysis(SMA), Linear Spectral unmixing

• SMA: determines the proportion or abundance of landcover per pixel

  • Assumption: the reflectance measured at each pixel is represented by the linear sum of endmembers weighted by the associated endmember fraction

• considerations:

  • pixel purity

  • number of endmembers

  • multiple endmember spectral analysis

7.1.3 Accuracy assessment

TP (True Positive): models predicts positive class correctly

TN (True Negative): models predicts negative class correctly

FP (False Positive): models predicts positive class incorrectly

FN (False Negative): models predicts negative class incorrectly

• PA (producer’s accuracy)=\(\frac{TP}{TP+FN}\)

• UA (user’s accuracy)=\(\frac{TP}{TP+FP}\)

• OA (the overall accuracy)=\(\frac{TP+TN}{TP+FP+FN+TN}\)

• Errors of omission = 100 - PA

• Errors of commission = 100 - UA

• kappa coefficient: to express the accuracy of an image compared to the results by chance

• F1 score: combines both recall(PA) and precision(UA)

  • F1 = \(\frac{TP+TN}{TP+\frac12*(FP+FN)}\),(0,1)

• Receiver Operating Characteristic Curve (the ROC Curve): to minimize noise from radar to identify true positives and not miss aircraft

  • Area Under the ROC Curve(AUC/AUROC): simply the area under the curve, could help us compare models easily.

• Train and Test Split: simply holding back a percentage of the original data used to train the model and then test it at the end.

• cross validation

  • leave one out cross validation: an extreme version of cross validation

    • use all the training data except 1 and repeats though all of it

  • spatial cross validation

    • Tobler’s first law: the near things are more related than distant things.

    • spatially partition the folded data, folds are from cross validation

7.2 Application

Pixel based analysis is appropriate for moderate and low resolution, however, as the spatial resolution of satellite images has become finer it makes difficult to classify at the pixel level(Jovanović et al. 2021). It relies on spectral values of individual pixels, is not able to capture the spatial context and variations of different land covers in high-resolution images(Campbell and Wynne 2011). OBIA as an alternative method could bypass the problem of pixel based analysis by by grouping a number of pixels into shapes with a meaningful representation of the objects.

Besides, OBIA has clear advantages in cooperating multi-source data, such as data obtained from Digital Terrain Maps (DTMs) and other thematic maps. This method also enhances classification tasks, especially in accurately mapping heterogeneous landscapes(Xofis and Poirazidis 2018).

Recent studies have been combining OBIA and Convolutional Neural Network (CNN), which has been shown to effectively extract high-level image features, maintain clear boundaries of image objects(Fu et al. 2019; Wang et al. 2018). In the research conducted by Tang, Li, and Wang (2020), they combined OBIA with CNN to form an object-based CNN to extract the tea plantations(sees Figure 1). They revealed that the use of CNN in OBIA setting can effectively improve mapping accuracy, particularly in terms of geometric accuracy.

Figure 1. The graphical abstract

7.3 Reflection

• A very important point in image classification is transferability especially in remote sensing and image analysis applications, where the data can have significant variations in terms of spatial resolution, spectral bands, and imaging sensors. In such cases, a classifier that can transfer well across different datasets and imaging conditions is essential to ensure reliable and accurate results.

Campbell, James B., and Randolph H. Wynne. 2011. Introduction to Remote Sensing, Fifth Edition. Guilford Press.

Fu, Yongyong, Kunkun Liu, Zhangquan Shen, Jinsong Deng, Muye Gan, Xinguo Liu, Dongming Lu, and Ke Wang. 2019. “Mapping Impervious Surfaces in TownRural Transition Belts Using China’s GF-2 Imagery and Object-Based Deep CNNs.” Remote Sensing 11 (3): 280. https://doi.org/10.3390/rs11030280.

Jovanović, Dušan, Milan Gavrilović, Dubravka Sladić, Aleksandra Radulović, and Miro Govedarica. 2021. “Building Change Detection Method to Support Register of Identified Changes on Buildings.” Remote Sensing 13 (16): 3150. https://doi.org/10.3390/rs13163150.

Tang, Zixia, Mengmeng Li, and Xiaoqin Wang. 2020. “Mapping Tea Plantations from VHR Images Using OBIA and Convolutional Neural Networks.” Remote Sensing 12 (18): 2935. https://doi.org/10.3390/rs12182935.

Wang, Lei, Yang Chen, Luliang Tang, Rongshuang Fan, and Yunlong Yao. 2018. “Object-Based Convolutional Neural Networks for Cloud and Snow Detection in High-Resolution Multispectral Imagers.” Water 10 (11): 1666. https://doi.org/10.3390/w10111666.

Xofis, Panteleimon, and Konstantinos Poirazidis. 2018. “Combining Different Spatio-Temporal Resolution Images to Depict Landscape Dynamics and Guide Wildlife Management.” Biological Conservation 218 (February): 10–17. https://doi.org/10.1016/j.biocon.2017.12.003.