This paper studies the use of deep-learning models (AlexNet, VggNet, ResNet) pre-trained on object categories (ImageNet) in applied texture classification problems such as plant disease detection tasks. Research related to precision agriculture is of high relevance due to its potential economic impact on agricultural productivity and quality. Within this context, we propose a deep learning-based feature extraction method for the identification of plant species and the classification of plant leaf diseases. We focus on results relevant to real-time processing scenarios that can be easily transferred to manned/unmanned agricultural smart machinery (e.g. tractors, drones, robots, IoT smart sensor networks, etc.) by reconsidering the common processing pipeline. In our approach, texture features are extracted from different layers of pre-trained Convolutional Neural Network models and are later applied to a machine-learning classifier. For the experimental evaluation, we used publicly available datasets consisting of RGB textured images and datasets containing images of healthy and non-healthy plant leaves of different species. We compared our method to feature vectors derived from traditional handcrafted feature extraction descriptors computed for the same images and end-to-end deep-learning approaches. The proposed method proves to be significantly more efficient in terms of processing times and discriminative power, being able to surpass traditional and end-to-end CNN-based methods and provide a solution also to the problem of the reduced datasets available for precision agriculture.

Lately, 3D imaging techniques have achieved a lot of progress due to recent developments in 3D sensor technologies. This leads to a great interest regarding 3D image feature extraction and classification techniques. As pointed out in literature, one of the most important and discriminative features in images is the textural content. Within this context, we propose a texture feature extraction technique for volumetric images with improved discrimination power. The method could be used in textured volumetric data classification tasks. To achieve this, we fuse two complementary pieces of information, feature vectors derived from Local Binary Patterns (LBP) and the Gray-Level Co-occurrence Matrix-based methods. They provide information regarding the image pattern and the contrast, homogeneity and local anisotropy in the volumetric data, respectively. The performance of the proposed technique was evaluated on a public dataset consisting of volumetric textured images affected by several transformations. The classifiers used are the Support Vector Machine, k-Nearest Neighbours and Random Forest. Our method outperforms other handcrafted 3D or 2D texture feature extraction methods and typical deep-learning networks. The proposed technique improves the discrimination power and achieves promising results even if the number of images per class is relatively small.

This paper presents an improved feature extraction method based on the use of state-of-the art filtering techniques and Local Binary Patterns-derived feature descriptors with applications in texture classification. The method is adaptive, being capable to determine the type of noise present in the input image and to apply the appropriate operator for the filtering step of the feature extraction technique. The improved approaches, labelled BM3DELBP (Block Matching and 3D filtering Extended Local Binary Pattern) and SARBM3DELBP (Synthetic Aperture Radar Block Matching and 3D filtering Extended Local Binary Pattern) bring significant improvements both in terms of robustness to Gaussian and speckle noise and in terms of classification accuracy, being invariant to different image transformations. We tested our approach both on synthetic textures from two standard Outex databases and on real polarimetric Synthetic Aperture Radar (SAR) images of pine forests. On all considered databases, the proposed approach proved to be above state-of-the-art LBP variants in terms of classification accuracy, even in the presence of high Gaussian and speckle noise levels.

This paper proposes new feature extractors for colour texture classification based on Local Binary Patterns (LBP) operators which are invariant to rotation, illumination and change of the observation scale, being also robust to Gaussian noise. Our novel approaches are operating in the RGB colour space and are using a state-of-the-art image denoising algorithm embedded into a multiscale LBP feature extraction process. The results obtained in the experimental section for grape leaf disease classification on a public database, using Support Vector Machines, show that the proposed feature extractors bring a significant improvement in terms of accuracy, precision and recall when compared to recent grayscale LBP-based approaches.

This paper proposes an improved feature extraction method for volumetric texture classification. Our approach consists in the computation of 3D co-occurrence matrices built using both the image intensity and the gradient image information. The feature vector represents the concatenation of the Haralick second-order statistics and the proposed gradient-based and orientation-based indicators. The results obtained on a public synthetic 3D texture database show that the proposed technique is more discriminative and brings improvements in the classification performance when compared to recent 3D and 2D texture descriptors.

This paper presents an extension of the Local Binary Patterns feature descriptors to hyperspectral image classification. Our approach uses a Principal Component Analysis technique to extract the most representative bands and for each pixel, it concatenates the histograms obtained for each selected band. The histograms are built from features which are discriminative and invariant to different transformations in the input. The proposed method achieves promising results for hyperspectral image classification. We tested our technique on four publicly available hyperspectral image databases of Earth observation images and for all of them the proposed method improved the classification accuracy when compared to the classical Local Binary Pattern approach.