Automatic detection of Ceratocystis wilt in Eucalyptus crops from aerial images

Ceratocystiswilt and canker – a disease that compromises the growing of commercialAcacia-based plantations in the tropics

“…To distinguish between healthy, infected and dead oak trees, the data was linked to leaf chlorophyll concentration using three spectral bands: visible green (0.52-0.60 µm), visible red (0.63-0.69 µm) and nearinfrared (0.63-0.69); infected trees had veinal necrosis and tip burn (Everitt et al 1999). In a second study (Souza et al 2015), a camera mounted in an unmanned aerial vehicle (UAV) detected wilt disease caused by C. fimbriata on Eucalyptus spp. The images were analysed and compared using four distinct machine learning techniques: K-Nearest Neighbours (K-NN), Random Forest (RF), Artificial Neural Network (ANN) and Gaussian Processes (GP).…”

Detecção de patologias em plantações de eucaliptos com aprendizado de máquina

“…O t reinament o do classi}cador ut ilizou caract eríst icas ext raídas a part ir de Blocos Cont ext uais de forma análoga à descrit a em Souza et al (2015), sendo est a met odologia t ambém empregada em out ros domínios, como na det ecção de vias navegáveis para veículos aut ônomos FRÉMONT;). Para criar a mat riz de caract erísticas, um Bloco Int erno de 4x4 pixels em uma janela deslizant e percorre a imagem e veri}ca se 100% dos pixels de cada amost ra são pert encent es a apenas uma única classe.…”

“…Nest e domínio, para os t est es com o classi}cador, o t amanho de Bloco Int erno foi }xado em 4x4 pixels pois a ut ilização de blocos int ernos maiores acarret aria em grande perda de exemplos de eucalipt os doent es. Para a ext ração das features das imagens foram ut ilizadas as bibliot ecas Scikit-Image (WALT et al, 2014) SOUZA et al, 2015).…”

Navigability estimation for autonomous vehicles using machine learning

“…We also evaluate how the road blocks affect the performance. For that purpose, we removed the road blocks features while maintaining the best radius parameter previous obtained (3) and all image features. Table 4 shows the results where the column "Diff."…”

“…To compare our method with others, we submitted our results, under the name of FCN-LC (Fully Convolutional Network -Large Context), to the KITTI road detection benchmark 3 . Table 9 presents the first nine results in the Urban Road category (at submission time) which is an aggregate of all three road categories (UU, UM, and UMM).…”