Adversarial Discriminative Domain Adaptation for Deforestation Detection

Abstract

Abstract. Although very efficient in a number of application fields, deep learning based models are known to demand large amounts of labeled data for training. Particularly for remote sensing applications, responding to that demand is generally expensive and time consuming. Moreover, supervised training methods tend to perform poorly when they are tested with a set of samples that does not match the general characteristics of the training set. Domain adaptation methods can be used to mitigate those problems, especially in applications where labeled data is only available for a particular region or epoch, i.e., for a source domain, but not for a target domain on which the model should be tested. In this work we introduce a domain adaptation approach based on representation matching for the deforestation detection task. The approach follows the Adversarial Discriminative Domain Adaptation (ADDA) framework, and we introduce a margin-based regularization constraint in the learning process that promotes a better convergence of the model parameters during training. The approach is evaluated using three different domains, which represent sites in different forest biomes. The experimental results show that the approach is successful in the adaptation of most of the domain combination scenarios, usually with considerable gains in relation to the baselines.

Domain-Adversarial Neural Networks for Deforestation Detection in Tropical Forests

“…The No Domain Adaptation row, regarded as the base-line classifier, corresponds to a classification scheme in which the Feature Extractor and the Label Predictor are trained on the source domain and evaluated directly on the target domain. The remaining rows show the results of the different domain adaptation methods, specifically: the original DANN method [8]; the ADDA-based approach [5]; the CycleGAN DN approach [6]; and the extension of DANN method proposed in this work (DANN+CVA). We observe that the mAP values correspond to the area under the curve obtained when computing pairs of Precision and Recall values for different classification thresholds, in the range of 0 to 1, over the average of probability maps delivered by each classifier.…”

“…The study areas were the same ones used in [5] and [6]. Two of the sites are located in the Amazon biome, specifically in the Brazilian states of Rondônia (RO) and Pará (PA).…”

“…Analogous to [5], [6], we split the images pairs from the different sites into non-overlapping tiles, from which we selected the training, validation and test samples/patches. We split the larger images from RO into 100 tiles and took approximately 20%, 5% and 75% for training, validation and testing, respectively.…”

“…To date, a few different DA approaches have been employed for CD on urban areas, e.g., [1], [2], [3], and [4]. To the best of our knowledge, only two DA approaches have focused on the deforestation detection task, namely: [5] and [6]. Such DA approaches can be categorized as representation matching and appearance adaptation techniques [7].…”

“…Appearance adaptation is mainly supported by so-called Image-to-Image (I2I) translation methods, which aim at transforming images from a target domain so that the generated images have a style that is similar to the images from a source domain. Representation matching aims at aligning features extracted from the domains in a common latent feature space, and involves either creating a different classifier for each domain [5], or a single classifier that is able to properly classify data from both domains [8].…”