Unsupervised image translation using adversarial networks for improved plant disease recognition

Abstract

Acquisition of data in task-specific applications of machine learning like plant disease recognition is a costly endeavor owing to the requirements of professional human diligence and time constraints. In this paper, we present a simple pipeline that uses GANs in an unsupervised image translation environment to improve learning with respect to the data distribution in a plant disease dataset, reducing the partiality introduced by acute class imbalance and hence shifting the classification decision boundary towards better performance. The empirical analysis of our method is demonstrated on a limited dataset of 2789 tomato plant disease images, highly corrupted with an imbalance in the 9 disease categories. First, we extend the state of the art for the GAN-based image-to-image translation method by enhancing the perceptual quality of the generated images and preserving the semantics. We introduce AR-GAN, where in addition to the adversarial loss, our synthetic image generator optimizes on Activation Reconstruction loss (ARL) function that optimizes feature activations against the natural image. We present visually more compelling synthetic images in comparison to most prominent existing models and evaluate the performance of our GAN framework in terms of various datasets and metrics. Second, we evaluate the performance of a baseline convolutional neural network classifier for improved recognition using the resulting synthetic samples to augment our training set and compare it with the classical data augmentation scheme. We observe a significant improvement in classification accuracy (+5.2%) using generated synthetic samples as compared to (+0.8%) increase using classic augmentation in an equal class distribution environment.

Assessing the potential for deep learning and computer vision to identify bumble bee species from images

“…For example, metric learning could be used to learn similar features between a pair of images, thus enhancing the discriminative power of deep CNNs 37 . Alternatively, generative adversarial networks 38 may help improve error rates for poorly sampled species and low accuracy due to class imbalance by generating synthetic image data when new images are difficult to acquire.…”

Convolutional Neural Networks for Image-Based High-Throughput Plant Phenotyping: A Review

“…Compared with studies in the first phase, the two pioneering studies demonstrated the importance of understanding the mechanism of CNNs for stress phenotyping as well as the potential for stress severity quantification. Image annotation is still recognized as a limiting factor for using many DL algorithms (especially supervised ones), so researchers investigated the use of generative adversarial networks (GANs) to generate synthetic images for training CNN models for plant stress detection and classification [ 54 ]. AR-GAN based on Cycle-GAN was developed to translate contextual information learned between different image sets.…”

Review of the State of the Art of Deep Learning for Plant Diseases: A Broad Analysis and Discussion

“…Autoregressive Deep Convolutional Generative Adversarial Network (AR-GAN) [ 130 ] is an approach based on three optimisation functions: the standard GAN, cycle self-consistency, and, to increase the affinity between generated and original images in terms of quality, reconstruction activation [ 131 , 132 ]. The created dataset has been shown to enhance classification by +5.2%.…”

“…It is important to note that there are both quantitative and qualitative methods of evaluation for GANs. Average precision is based upon a comparison between the label maps of generated and the real images using semantic segmentation metrics (intersection-over-union per pixel/class).The Fréchet inception distance tool measures the covariance of feature distributions in the real/generated data [ 133 ]; the computational efficiency of this method has also been proven for large-scale datasets [ 130 ]. Precision measures the quality of created images compared to the corresponding learned dataset, while recall measures the diversity of the generated data [ 133 ].…”