Convolutional neural networks (CNNs) have emerged as one of the most popular ways applied in many fields. These networks deliver better performance when going deeper and larger. However, the complicated computation and huge storage impede hardware implementation. To address the problem, quantized networks are proposed. Besides, various convolutional structures are designed to meet the requirements of different applications. For example, compared with the traditional convolutions (CONVs) for image classification, CONVs for image generation are usually composed of traditional CONVs, dilated CONVs, and transposed CONVs, leading to a difficult hardware mapping problem. In this brief, we translate the difficult mapping problem into the sparsity problem and propose an efficient hardware architecture for sparse binary and ternary CNNs by exploiting the sparsity and low bit-width characteristics. To this end, we propose an ineffectual data removing (IDR) mechanism to remove both the regular and irregular sparsity based on dual-channel processing elements (PEs). Besides, a flexible layered load balance (LLB) mechanism is introduced to alleviate the load imbalance. The accelerator is implemented with 65-nm technology with a core size of 2.56 mm^2. It can achieve 3.72-TOPS/W energy efficiency at 50.1 mW, which makes it a promising design for embedded devices.
Over the past four decades, China has invested $298 billion in cultivated land consolidation (CLC), making a substantial contribution to increased agricultural production and enhanced food security despite rapid urban expansion and cultivated land encroachment. However, in the absence of a quantitative, spatial evaluation of CLC potential that engages local stakeholders including farmers, community, and decision-makers, the scope for systematic planning to guide and influence the prioritization, effectiveness, and efficiency of CLC for achieving enduring benefits for agricultural productivity and sustainability is limited. In this study, we used a knowledge co-production approach with spatial multi-criteria analysis (MCA) for evaluating CLC potential in Jiangsu Province and identifying priority areas with the participation of stakeholders. We identified key priority areas and considerations for ensuring CLC implementation for supporting the transition to more productive and sustainable agricultural systems including 7 priority types across 813 townships. 41.82 % of the areas were targeted for resource consolidation, 33.83 % for pattern consolidation, and 9.84 % for utilization consolidation, with the remaining 14.51% for combined consolidation based on multiple criteria. While 686 townships with low and very low CLC potential were incorporated into the CLC planning as protection and development priority types. These spatially explicit results can inform regional prioritization and targeting investment in CLC projects with a high level of buy-in from stakeholders. The methodology and findings of this study can inform innovation in systematic spatial planning with stakeholders for CLC across China and help manage the trade-offs between urbanization and food security.
Decoupling the drivers of cropland productivity is critical for food security, yet the non-transferability of mechanisms across spatial scales hinders consensus. Productivity is an emergent property of a social-ecological system (SES). Here, we deconstruct the drivers of cropland productivity change across China (2001–2020) using a multi-scale analytical framework that integrates machine learning and causal inference. Our analysis reveals three key findings. First, productivity is co-determined by a nexus of biophysical properties (24.2% relative importance), urbanization (17.4%), and cropland-use intensity (16.2%), with pronounced non-linear interactions. Second, the dominance of these drivers is scale-dependent, shifting systematically from macro-scale biophysical processes to micro-scale socioeconomic behaviours. Third, external macro-scale disturbances primarily exert their influence through indirect micro-scale pathways; for instance, urbanization’s main impact is not direct land conversion, but the suppression of cropland use intensification, making it the dominant negative driver in 89% of counties. These findings demonstrate that cropland productivity dynamics are governed by a network of multi-scale, indirect causal mechanisms. Effective governance requires a systems-based approach that moves beyond single-factor interventions to manage these complex interactions.