A Low-Cost, Low-Power and Real-Time Image Detector for Grape Leaf Esca Disease Based on a Compressed CNN

Esca is one of the most common grape leaf diseases that seriously affect grape yield, causing a loss of global production in the range of 20%–40%. Therefore, a timely and effective identification of the disease could help to develop an early treatment approach to control its spread while reducing economic losses. For this purpose the use of computer vision and machine learning techniques for recognizing plant diseases have been extensively studied in recent years. The aim of this paper is to propose an image detector based on a high-performance convolutional neural network (CNN) implemented in a low cost, low power platform, to monitor the Esca disease in real-time. To meet the severe constraints typical of an embedded system, a new low-rank CNN architecture (LR-Net) based on CANDECOMP/PARAFAC (CP) tensor decomposition has been developed. The compressed CNN network so obtained has been trained on a specific dataset and implemented in a low-power, low-cost Python programmable machine vision camera for real-time classification. An extensive experimentation has been conducted and the results achieved show the superiority of LR-Net with respect to the state-of-the-art networks both in terms of inference time and memory occupancy.

I. Introduction

The vineyards have different degrees of internal variability [1], [2] due to the complex interactions that are established among the morphology of the soil, its chemical and biological fertility, the seasonal meteorological trends and the management techniques applied. These interactions influence the physiological performance of the grapevines, changing the balance between the vegetative and productive components [3], [4]. Thus, the conditions for the vegeto-productive imbalance in plants may arise, negatively affecting the production of good quality grapes and the resilience to biotic and abiotic stresses. The monitoring of spatial and/or temporal variability within vineyards was used to calibrate water [5] or nitrogen requirements [6], to define new protocols of sampling grapes for maturation monitoring [7] and lately to detect foliar symptoms of the Esca grapevine trunk disease [8]. Esca is one of the main grapevine diseases, able to induce severe symptoms and lead to the death of symptomatic plants worldwide [9], [10]. A list of fungal pathogens have been associated with the disease, and the most common are Phaeomoniella chlamydospora, Phaeoacremonium ultimum and Fomitiporia mediterranea [11]. External disease symptoms, including leaf stripes, interveinal chlorosis and necrosis, linear pitting on berries, and dessiccation of a portion or of the whole plant appears on the canopy, even in young vineyards few years after planting, and become increasingly severe with the aging of the plant [12]. Infected plants do not show foliar symptoms every years, and investigations run in the Marche region (Italy) on Verdicchio have indicated that around one third of plants showed symptoms three years out of three, one third two years out of three, and the last third just one year out of three. Moreover, the symptoms displayed are highly variable according to the cultivator and are affected by the rootstock [13], [14]. Therefore, having tools that are useful for the grower to monitor the progress of the disease in real-time every year can help to explain the correlation with other factors and apply mitigation strategies.