Optimised U-Net for Land Use–Land Cover Classification Using Aerial Photography
Convolutional Neural Networks (CNN) consist of various hyper-parameters which need to be specifed or can be altered when defning a deep learning architecture. There are numerous studies which have tested diferent types of networks (e.g. U-Net, DeepLabv3+) or created new architectures, benchmarked against well-known test datasets. However, there is a lack of real-world mapping applications demonstrating the efects of changing network hyper-parameters on model performance for land use and land cover (LULC) semantic segmentation. In this paper, we analysed the efects on training time and classifcation accuracy by altering parameters such as the number of initial convolutional flters, kernel size, network depth, kernel initialiser and activation functions, loss and loss optimiser functions, and learning rate. We achieved this using a well-known top performing architecture, the U-Net, in conjunction with LULC training data and two multispectral aerial images from North Queensland, Australia. A 2018 image was used to train and test CNN models with diferent parameters and a 2015 image was used for assessing the optimised parameters. We found more complex models with a larger number of flters and larger kernel size produce classifcations of higher accuracy but take longer to train. Using an accuracy-time ranking formula, we found using 56 initial flters with kernel size of 5×5 provide the best compromise between training time and accuracy. When fully training a model using these parameters and testing on the 2015 image, we achieved a kappa score of 0.84. This compares to the original U-Net parameters which achieved a kappa score of 0.73.
Pre-Processing Training Data Improves Accuracy and Generalisability of Convolutional Neural Network Based Landscape Semantic Segmentation
Data pre-processing for developing a generalised land use and land cover (LULC) deep learning model using earth observation data is important for the classification of a different date and/or sensor. However, it is unclear how to approach deep learning segmentation problems in earth observation data. In this paper, we trialled different methods of data preparation for Convolutional Neural Network (CNN) training and semantic segmentation of LULC features within aerial photography over the Wet Tropics and Atherton Tablelands, Queensland, Australia. This was conducted by trialling and ranking various training patch selection sampling strategies, patch and batch sizes, data augmentations and scaling and inference strategies. Our results showed: a stratified random sampling approach for producing training patches counteracted class imbalances" a smaller number of larger patches (small batch size) improves model accuracy" data augmentations and scaling are imperative in creating a generalised model able to accurately classify LULC features in imagery from a different date and sensor" and producing the output classification by averaging multiple grids of patches and three rotated versions of each patch produced a more accurate and aesthetic result. Combining the findings from the trials, we fully trained five models on the 2018 training image and applied the model to the 2015 test image. The output LULC classifications achieved an average kappa of 0.84, user accuracy of 0.81, and producer accuracy of 0.87. Future research using CNNs and earth observation data should implement the findings of this project to increase LULC model accuracy and transferability.
Alternatives to Landsat-5 Thematic Mapper for operational monitoring of vegetation cover: considerations for natural resource management agencies
La possibilite´ que le capteur Thematic Mapper (TM) de Landsat 5 ne puisse fournir des images jusqu'au moment du lancement de la mission LDCM (« Landsat Data Continuity Mission ») pre´vu pour de´cembre 2012 suscite des inquie´tudes. La crainte vient du manque de combustible et des proble`mes sporadiques du satellite et du capteur. Le mauvais fonctionnement du correcteur de lignes de balayage (SLC) du capteur ETM+ (« Enhanced Thematic Mapper ») de Landsat 7 en 2003 fait que les images de ETM+ renferment des bandes de donne´es manquantes. Conse´quemment, on craint que les images de qualite´ supe´rieure de Landsat ne soient pas disponibles pour fins de suivi durant les anne´es 2011 et 2012. Les agences de suivi des ressources naturelles font ainsi face au de´fi conside´rable de de´terminer une alternative ade´quate aux images de TM de Landsat 5 en cas de besoin. Bien qu'il existe une litte´rature sur les alternatives possibles, il y a peu d'e´tudes qui donnent des directives aux agences de suivi sur les proble´matiques a` explorer lorsque l'on conside`re une source de donne´es alternatives. On a entrepris une e´tude qui a permis d'identifier les donne´es de ETM+ de Landsat 7, HRVIR de SPOT 4, HRG de SPOT 5 et de LISS-III de IRS-P6 comme e´tant les alternatives les mieux adapte´es pour le suivi du couvert de ve´ge´tation dans le Queensland et la Nouvelle-Galles du Sud (NSW) dans l'est de l'Australie. On a trouve´ qu'aucun des capteurs n'e´tait ide´al a` cause d'une combinaison de l'un ou de plusieurs des facteurs suivants dont la qualite´ radiome´trique re´duite, les volumes de donne´es accrus, les besoins de traitement supple´mentaires et les couˆ ts d'acquisition plus e´leve´s. On a trouve´ que le faible couˆ t et la facilite´ d'acce`s des donne´es de ETM+ de Landsat 7 rendaient cette option plus viable aux plans technique et e´conomique pour le suivi annuel de l'e´tendue et deschangements de la ve´ge´tation ligneuse dans l'est de l'Australie. Cependant, ETM+ de Landsat 7 ne peut eˆtre conside´re´ comme une option dans plusieurs re´gions du monde duˆ a` l'importance du couvert nuageux. Nous avons utilise´ les expe´riences acquises durant ce travail pour recommander une proce´dure que les agences de suivi des ressources naturelles peuvent mettre de l'avant pour explorer les proble´matiques relie´es au choix d'une alternative aux donne´es de TM de Landsat 5 pour le suivi du couvert.