Search Results
An Assessment of Deep Learning Models and Word Embeddings for Toxicity Detection within Online Textual Comments
2021, Dessì, Danilo, Recupero, Diego Reforgiato, Sack, Harald
Today, increasing numbers of people are interacting online and a lot of textual comments are being produced due to the explosion of online communication. However, a paramount inconvenience within online environments is that comments that are shared within digital platforms can hide hazards, such as fake news, insults, harassment, and, more in general, comments that may hurt someone’s feelings. In this scenario, the detection of this kind of toxicity has an important role to moderate online communication. Deep learning technologies have recently delivered impressive performance within Natural Language Processing applications encompassing Sentiment Analysis and emotion detection across numerous datasets. Such models do not need any pre-defined hand-picked features, but they learn sophisticated features from the input datasets by themselves. In such a domain, word embeddings have been widely used as a way of representing words in Sentiment Analysis tasks, proving to be very effective. Therefore, in this paper, we investigated the use of deep learning and word embeddings to detect six different types of toxicity within online comments. In doing so, the most suitable deep learning layers and state-of-the-art word embeddings for identifying toxicity are evaluated. The results suggest that Long-Short Term Memory layers in combination with mimicked word embeddings are a good choice for this task.
A Pixel-wise Segmentation Model to Identify Bur Chervil (Anthriscus caucalis M. Bieb.) Within Images from a Cereal Cropping Field
2022, Karimi, Hadi, Navid, Hossein, Dammer, Karl-Heinz
Because of insufficient effectiveness after herbicide application in autumn, bur chervil (Anthriscus caucalis M. Bieb.) is often present in cereal fields in spring. A second reason for spreading is the warm winter in Europe due to climate change. This weed continues to germinate from autumn to spring. To prevent further spreading, a site-specific control in spring is reasonable. Color imagery would offer cheap and complete monitoring of entire fields. In this study, an end-to-end fully convolutional network approach is presented to detect bur chervil within color images. The dataset consisted of images taken at three sampling dates in spring 2018 in winter wheat and at one date in 2019 in winter rye from the same field. Pixels representing bur chervil were manually annotated in all images. After a random image augmentation was done, a Unet-based convolutional neural network model was trained using 560 (80%) of the sub-images from 2018 (training images). The power of the trained model at the three different sampling dates in 2018 was evaluated at 141 (20%) of the manually annotated sub-images from 2018 and all (100%) sub-images from 2019 (test images). Comparing the estimated and the manually annotated weed plants in the test images the Intersection over Union (Jaccard index) showed mean values in the range of 0.9628 to 0.9909 for the three sampling dates in 2018, and a value of 0.9292 for the one date in 2019. The Dice coefficients yielded mean values in the range of 0.9801 to 0.9954 for 2018 and a value of 0.9605 in 2019.
Deep Learning Object Detection for Image Analysis of Cherry Fruit Fly (Rhagoletis cerasi L.) on Yellow Sticky Traps
2022, Salamut, Christian, Kohnert, Iris, Landwehr, Niels, Pflanz, Michael, Schirrmann, Michael, Zare, Mohammad
Insect populations appear with a high spatial, temporal and type-specific diversity in orchards. One of the many monitoring tools for pest management is the manual assessment of sticky traps. However, this type of assessment is laborious and time-consuming so that only a few locations can be controlled in an orchard. The aim of this study is to test state-of-the art object detection algorithms from deep learning to automatically detect cherry fruit flies (Rhagoletis cerasi), a common insect pest in cherry plantations, within images from yellow sticky traps. An image annotation database was built with images taken from yellow sticky traps with more than 1600 annotated cherry fruit flies. For better handling in the computational algorithms, the images were augmented to smaller ones by the known image preparation methods “flipping” and “cropping” before performing the deep learning. Five deep learning image recognition models were tested including Faster Region-based Convolutional Neural Network (R-CNN) with two different methods of pretraining, Single Shot Detector (SSD), RetinaNet, and You Only Look Once version 5 (YOLOv5). R‑CNN and RetinaNet models outperformed other ones with a detection average precision of 0.9. The results indicate that deep learning can act as an integral component of an automated system for high-throughput assessment of pest insects in orchards. Therefore, this can reduce the time for repetitive and laborious trap assessment but also increase the observed amount of sticky traps
This looks More Like that: Enhancing Self-Explaining Models by prototypical relevance propagation: This Looks More Like That
2022, Gautam, Srishti, Höhne, Marina M.-C., Hansen, Stine, Jenssen, Robert, Kampffmeyer, Michael
Current machine learning models have shown high efficiency in solving a wide variety of real-world problems. However, their black box character poses a major challenge for the comprehensibility and traceability of the underlying decision-making strategies. As a remedy, numerous post-hoc and self-explanation methods have been developed to interpret the models’ behavior. Those methods, in addition, enable the identification of artifacts that, inherent in the training data, can be erroneously learned by the model as class-relevant features. In this work, we provide a detailed case study of a representative for the state-of-the-art self-explaining network, ProtoPNet, in the presence of a spectrum of artifacts. Accordingly, we identify the main drawbacks of ProtoPNet, especially its coarse and spatially imprecise explanations. We address these limitations by introducing Prototypical Relevance Propagation (PRP), a novel method for generating more precise model-aware explanations. Furthermore, in order to obtain a clean, artifact-free dataset, we propose to use multi-view clustering strategies for segregating the artifact images using the PRP explanations, thereby suppressing the potential artifact learning in the models.
Improving Deep Learning-based Plant Disease Classification with Attention Mechanism
2022, Alirezazadeh, Pendar, Schirrmann, Michael, Stolzenburg, Frieder
In recent years, deep learning-based plant disease classification has been widely developed. However, it is challenging to collect sufficient annotated image data to effectively train deep learning models for plant disease recognition. The attention mechanism in deep learning assists the model to focus on the informative data segments and extract the discriminative features of inputs to enhance training performance. This paper investigates the Convolutional Block Attention Module (CBAM) to improve classification with CNNs, which is a lightweight attention module that can be plugged into any CNN architecture with negligible overhead. Specifically, CBAM is applied to the output feature map of CNNs to highlight important local regions and extract more discriminative features. Well-known CNN models (i.e. EfficientNetB0, MobileNetV2, ResNet50, InceptionV3, and VGG19) were applied to do transfer learning for plant disease classification and then fine-tuned by a publicly available plant disease dataset of foliar diseases in pear trees called DiaMOS Plant. Amongst others, this dataset contains 3006 images of leaves affected by different stress symptoms. Among the tested CNNs, EfficientNetB0 has shown the best performance. EfficientNetB0+CBAM has outperformed EfficientNetB0 and obtained 86.89% classification accuracy. Experimental results show the effectiveness of the attention mechanism to improve the recognition accuracy of pre-trained CNNs when there are few training data.